Proving Copyright Infringement: Access, Substantial Similarity, and the Tests Courts Actually Use

Summary. A copyright plaintiff must prove two things: ownership of a valid copyright and copying of protectable expression, and the second element splits into copying in fact and unlawful appropriation. This article walks through each element in the order a court takes them, from the originality threshold of Feist and the registration prerequisite of Fourth Estate through access, probative similarity, striking similarity, and the Ninth Circuit's abandonment of the inverse ratio rule in Skidmore v. Led Zeppelin. It then explains the circuit tests for substantial similarity in real detail: the Second Circuit's ordinary observer and more discerning observer tests, the Ninth Circuit's extrinsic and intrinsic tests, and the abstraction-filtration-comparison method courts apply to software. It covers the filters that remove ideas, facts, merger, scenes a faire, and thin compilations from the comparison, and it shows how those filters decided the leading music, literary, visual, and software cases. The article closes with the role of experts and surveys, summary judgment and jury instruction practice, the economics of these cases, strategy for each side, a worked example, and an FAQ.

A songwriter hears a chorus on the radio that sounds like the demo she sent to a producer two years ago. A game studio notices that a competitor's new title has the same level structure, the same inventory screen, and a bug that only its own code ever had. A novelist watches a streaming series whose second season tracks her unpublished manuscript beat for beat. Each of them is certain they have been copied, and each of them is about to learn that certainty is not an element of the claim.

Copyright infringement is one of the few areas of law where the thing everyone argues about (do these two works look, sound, or read alike?) is not actually the question. The question is whether the defendant copied, and whether what was copied is the kind of thing copyright protects. Two works can be nearly identical and not infringe; two works can look quite different and infringe badly. This article explains how courts sort one from the other.

The short answer

To win an infringement claim, a plaintiff must prove (1) ownership of a valid copyright and (2) copying of constituent elements of the work that are original. That formulation comes from Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991), and every circuit uses some version of it.

The second element has two parts that courts routinely blur and that you must keep separate. Copying in fact (also called actual copying or factual copying) asks whether the defendant actually used the plaintiff's work rather than creating independently. It is proved by direct evidence, which is rare, or circumstantially through access plus similarities that are probative of copying, or through similarity so striking that copying is the only plausible explanation. Unlawful appropriation (also called improper appropriation or, confusingly, substantial similarity) asks whether the defendant took enough protectable expression to matter. Here the court filters out ideas, facts, standard elements, and anything the plaintiff did not originate, then compares what remains.

Three practical consequences follow. First, a plaintiff with no evidence of access and no striking similarity loses before anyone compares the works. Second, a defendant who copied verbatim can still win if what it copied was unprotectable. Third, the test for the second step varies by circuit in ways that change who decides (judge or jury), what evidence is admissible, and when the case can be dismissed. Those differences are the subject of most of this article.

The legal framework: what the statute says and what it does not

The Copyright Act grants the owner a bundle of exclusive rights listed in 17 U.S.C. § 106: to reproduce the work, prepare derivative works, distribute copies, and perform or display the work publicly. 17 U.S.C. § 501(a) makes anyone who violates one of those rights an infringer, and § 501(b) lets the legal or beneficial owner of an exclusive right sue for it.

What the statute does not do is define copying or say how much copying is too much. The words "substantial similarity" appear nowhere in Title 17. The entire apparatus of access, probative similarity, ordinary observers, and extrinsic tests is judge-made, which is why it varies by circuit and why the treatises matter. The leading discussion remains 4 Nimmer on Copyright § 13.03, which coined the "probative similarity" label most courts now use for the first step, and Patry on Copyright chapter 9 and Goldstein on Copyright § 9.1 offer competing frameworks that judges cite when the case law runs out.

Two statutory provisions do most of the work on the protectability side. 17 U.S.C. § 102(a) extends protection to "original works of authorship fixed in any tangible medium of expression." Section 102(b) then takes back a great deal: "In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work." Every filtration doctrine discussed below is an application of that sentence. 17 U.S.C. § 103 adds that protection for a compilation or derivative work extends only to the material contributed by the author, not to preexisting material.

The definitions in 17 U.S.C. § 101 matter more than they look. "Copies" are material objects in which a work is fixed; "derivative work" means a work based on one or more preexisting works; "compilation" means a work formed by the selection, coordination, or arrangement of preexisting materials. Whether a defendant's work is a copy, a derivative work, or an independent compilation that happens to overlap often decides which exclusive right is in play and how the comparison is framed.

Element one: ownership of a valid copyright

Originality, and how little of it is required

Feist is the case to read first. Rural Telephone published a white pages directory; Feist copied the listings. The Supreme Court held that facts are not copyrightable, that a compilation of facts is protected only to the extent its selection, coordination, or arrangement is original, and that originality requires independent creation plus "at least some minimal degree of creativity." Alphabetical listing of every subscriber in a service area did not clear even that low bar. The Court was explicit that this is a constitutional requirement, not a policy choice, and it rejected the "sweat of the brow" theory under which effort alone earned protection.

The practical lesson cuts both ways. The threshold is low: the Court said the "vast majority of works make the grade quite easily," and photographs qualify on the strength of choices about pose, lighting, and timing that date back to Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884). But the threshold is real, and a work that clears it barely gets a correspondingly thin copyright. A plaintiff whose work is a database, a form, a standard-shaped product, or a conventional arrangement of public domain material should expect the defendant to make originality the whole case.

The registration certificate and the presumption of validity

A certificate of registration made before or within five years after first publication is prima facie evidence of the validity of the copyright and of the facts stated in the certificate. 17 U.S.C. § 410(c). The presumption shifts the burden of production to the defendant, who can rebut it with evidence that the work is unoriginal, copied from a third party, or owned by someone else. A certificate obtained more than five years after publication gets whatever weight the court chooses to give it.

Defendants attack certificates on two fronts. The first is substantive: the work is not original, or the claimant is not the author or a proper transferee. The second is procedural: the application contained inaccurate information. Under 17 U.S.C. § 411(b), an inaccuracy invalidates the registration only if the applicant knew it was inaccurate and the Register would have refused registration had it known. The Supreme Court held in Unicolors, Inc. v. H&M Hennes & Mauritz, L.P., 595 U.S. 178 (2022) that "knowledge" means actual knowledge of the inaccuracy, and that a good-faith mistake of law (there, misunderstanding the "single unit of publication" rule for registering multiple fabric designs together) excuses an inaccuracy just as a good-faith mistake of fact does. The defense survives against an applicant who actually knew, or was willfully blind to, the error, and the Court noted that circumstantial evidence such as the applicant's experience and the clarity of the rule can support that finding. Before invalidating anything, the court must ask the Register whether the inaccuracy would have led to refusal, a referral that § 411(b)(2) makes mandatory.

Chain of title

A plaintiff who is not the author must prove a transfer. A transfer of copyright ownership is not valid unless it is in a signed writing. 17 U.S.C. § 204(a). Ownership vests initially in the author under 17 U.S.C. § 201, and for a work made for hire the employer or commissioning party is the author. Infringement cases are lost at this step more often than people expect: the "assignment" was an unsigned email, the contractor who built the website never transferred anything, the band never agreed who wrote the song. We cover these problems at length in Copyright Ownership, Joint Authorship, and Termination of Transfers and the Work Made for Hire Determination Checklist: A Practical Checklist; for this article, assume ownership is proven and note that a defendant should never assume it.

The registration prerequisite: Fourth Estate and § 412

For United States works, no civil action for infringement may be instituted until registration "has been made." 17 U.S.C. § 411(a). The Supreme Court resolved a long circuit split in Fourth Estate Public Benefit Corp. v. Wall-Street.com, LLC, 586 U.S. 296 (2019), holding that registration is made when the Copyright Office registers the claim (or refuses it), not when the applicant submits a complete application. A plaintiff who files suit on a pending application will be dismissed, usually without prejudice, and the fix is to register and refile.

The timing of registration also controls remedies. Under 17 U.S.C. § 412, statutory damages and attorney's fees are unavailable for infringement of an unpublished work that began before registration, or for infringement of a published work that began after publication and before registration, unless registration was made within three months of first publication. Because actual damages in a similarity case are often modest and hard to prove, § 412 frequently determines whether a case is worth bringing at all. Statutory damages under 17 U.S.C. § 504(c) run from $750 to $30,000 per work, up to $150,000 for willful infringement, and fees under 17 U.S.C. § 505 are discretionary either way. The registration mechanics are in Copyright Registration: A Comprehensive Guide, and the remedies in Copyright Damages and Remedies: Statutory Damages, Profits, Injunctions, and Fees.

One more timing rule: the three-year limitations period in 17 U.S.C. § 507(b) runs separately for each infringing act, so a plaintiff who waited a decade can still sue for the last three years of copying, as Petrella v. Metro-Goldwyn-Mayer, Inc., 572 U.S. 663 (2014) held in rejecting laches as a bar to damages within the statutory window.

Element two, part one: copying in fact

Where the two-step structure comes from

The modern structure traces to Arnstein v. Porter, 154 F.2d 464 (2d Cir. 1946), in which a prolific and litigious songwriter accused Cole Porter of stealing his melodies. Judge Jerome Frank, reversing summary judgment for Porter, separated two questions: whether the defendant copied from the plaintiff's work, and if so whether the copying went so far as to constitute improper appropriation. On the first question, he said, "analysis" and "dissection" are appropriate and expert testimony may help. On the second, the test is the response of the ordinary lay listener, and dissection and experts are irrelevant. Nearly every circuit test still carries that DNA, including the odd rule that an expert may testify about the first step but often not the second.

Direct evidence

Direct evidence of copying exists and is worth looking for: an admission in an email, a deposition concession, a witness who saw the defendant working from the plaintiff's file, a defendant's work that contains the plaintiff's watermark, a planted "trap" entry in a database, or a comment in source code that references the plaintiff's product. Software cases produce direct evidence more often than other kinds because version control systems, build logs, and developer chat preserve a record that creative industries do not. If you have it, lead with it; the access and probative similarity inquiry exists only because most plaintiffs do not.

Access: a reasonable possibility, not a bare possibility

Access means the defendant had a reasonable opportunity to view or hear the plaintiff's work. Courts phrase the standard as a "reasonable possibility," and they are emphatic that a "bare possibility" built on speculation is not enough. The Ninth Circuit's statement in Three Boys Music Corp. v. Bolton, 212 F.3d 477 (9th Cir. 2000) is the one most often quoted: access may be shown by (1) a chain of events linking the plaintiff's work to the defendant, such as submission to a publisher or producer who dealt with the defendant, or (2) wide dissemination of the plaintiff's work, such as a hit record or a best-selling book. In Three Boys the jury found that Michael Bolton, who grew up listening to the radio in Connecticut, had access to the Isley Brothers' 1966 single "Love Is a Wonderful Thing," which had charted modestly, and the Ninth Circuit affirmed over a vigorous argument that the evidence was thin.

The chain-of-events cases turn on the strength of the link. Sending a manuscript to a company is not access by everyone at the company; the plaintiff must show a nexus between the recipient and the people who created the accused work. Jorgensen v. Epic/Sony Records, 351 F.3d 46 (2d Cir. 2003), rejected "bare corporate receipt" where unsolicited demo tapes reached a label's mail room but no one connected them to the writers of the accused songs. Gaste v. Kaiserman, 863 F.2d 1061 (2d Cir. 1988), went the other way on a longer but more specific chain: the French song "Pour Toi" had been sent to a publisher who had a business relationship with the composer of "Feelings," and the jury was entitled to find access.

The wide-dissemination cases have moved online, and the Ninth Circuit's decision in Woodland v. Hill, 136 F.4th 1199 (9th Cir. 2025) shows the limits. A model and artist alleged that Lil Nas X copied poses from photographs posted to the plaintiff's Instagram account. The court affirmed dismissal: posts that drew between eight and seventy-five likes were not widely disseminated, the plaintiff alleged no chain of events connecting the defendant to the posts, and the fact that both parties used the same platform did not supply access. Public availability on the internet is not, by itself, access.

Probative similarity

Once access is shown, the plaintiff must also show similarities that tend to prove copying rather than coincidence. This is "probative similarity," a term courts adopted from Nimmer to distinguish it from the later question of substantial similarity of protectable expression. The distinction matters because at this step everything counts, including unprotectable elements. Shared errors, shared public domain material used in the same idiosyncratic way, a misspelling carried from one work to the other, the same fictitious street in a map, the same unexplained dead code in a program: none of these is protectable, but all of them are powerful evidence that one work was made from the other.

The Second Circuit adopted the label to clear up what it called the "considerable confusion" caused by using "substantial similarity" at two different points in the analysis: probative similarity, usually together with access, shows that copying happened, and substantial similarity then shows that the copying was unlawful. Laureyssens v. Idea Group, Inc., 964 F.2d 131 (2d Cir. 1992); Repp v. Webber, 132 F.3d 882 (2d Cir. 1997). Keep the two separate in every brief, every expert report, and every jury instruction. A list of shared features that are all unprotectable can be devastating proof of copying and still lose the case at the second step.

Striking similarity

Where access cannot be shown, the plaintiff may still prove copying by showing that the works are so strikingly similar that independent creation, coincidence, or a common source are all implausible. Selle v. Gibb, 741 F.2d 896 (7th Cir. 1984) is the leading case and a cautionary one. A Chicago songwriter claimed the Bee Gees copied "How Deep Is Your Love" from his unpublished song "Let It End." His expert testified the similarities were striking; the jury agreed; the trial judge granted judgment notwithstanding the verdict, and the Seventh Circuit affirmed. Striking similarity, the court held, means similarity "of a type which will preclude any explanation other than that of copying." The plaintiff's expert, a distinguished classical musicologist who had never before compared two popular songs, ruled out only independent creation. He never addressed coincidence or a prior common source (including the Bee Gees' own earlier songs), and he offered nothing on how complex or unique the shared passages were in a genre of short songs built on repeated themes. The court added that a plaintiff must still show some reasonable possibility of access, and Selle could not: he had played the song two or three times with his band in the Chicago area and mailed a tape to eleven record and publishing companies, eight of which sent it back.

Striking similarity can also keep a thin access case alive. In Repp, there was "little, if any, evidence demonstrating access" to a liturgical composer's song "Till You," but two musicologists swore that Andrew Lloyd Webber's "Phantom Song" was so similar to it as to preclude separate creation. The Second Circuit reversed summary judgment for Lloyd Webber because the district court had weighed the competing expert opinions and relied on its own listening, resolving a disputed fact instead of identifying it for trial.

Striking similarity does not dispense with the second step. A defendant who copied the unprotectable parts of a work has still not infringed. Ty, Inc. v. GMA Accessories, Inc., 132 F.3d 1167 (7th Cir. 1997), illustrates the doctrine working as intended: two plush pigs were so alike, down to features that had nothing to do with real pigs, that Judge Posner found copying could be inferred without proof of access, and because a plush pig can be designed many ways, the similarity also satisfied the second step.

The inverse ratio rule and its end

For decades the Ninth Circuit applied an "inverse ratio rule": the stronger the evidence of access, the less similarity the plaintiff needed to show. The rule had a plausible origin in the probative-similarity step, where strong access does make modest similarity more suspicious, but courts applied it unevenly and sometimes to the unlawful-appropriation step, where it made no sense (a defendant's opportunity to copy says nothing about whether what it took was protectable). The Second Circuit had rejected the rule in 1961 as one that "confuses more than it clarifies," Arc Music Corp. v. Lee, 296 F.2d 186 (2d Cir. 1961), and the Seventh Circuit in Peters v. West, 692 F.3d 629 (7th Cir. 2012) declined to endorse the idea that a high degree of access lowers the proof of similarity required, affirming dismissal of a claim that Kanye West's "Stronger" copied an earlier song of the same name even though the complaint plausibly alleged access.

The en banc Ninth Circuit abandoned the rule in Skidmore v. Led Zeppelin, 952 F.3d 1051 (9th Cir. 2020) (en banc), the "Stairway to Heaven" case. The trustee of the late Randy Wolfe (Randy California of the band Spirit) claimed that the opening of "Stairway" copied Spirit's instrumental "Taurus." The jury found for Led Zeppelin, a panel ordered a new trial over jury instructions, and the en banc court reinstated the verdict. Three holdings matter here. First, the inverse ratio rule is overruled: access and similarity are separate inquiries and access does not lower the similarity threshold. Second, for a work registered under the 1909 Act, the deposit copy (the sheet music filed with the Copyright Office) defines the scope of the copyright, so the jury properly compared "Stairway" to the one-page transcription of "Taurus" rather than to Spirit's recording. Third, the trial court did not err in declining a selection-and-arrangement instruction where the plaintiff had not presented that theory to the jury in a form that identified which unprotectable elements were combined in an original way. Because the decision is en banc, it binds every court in the circuit, and access now does one job there: it makes copying in fact more or less plausible. It never lowers the bar for the second step.

Independent creation and subconscious copying

Independent creation is a complete defense, and strictly speaking it is not a defense at all but a failure of proof on copying in fact. A defendant who can document that it created its work before it could have seen the plaintiff's, or who can produce drafts, sketches, session files, or commit histories showing a creative process that never touched the plaintiff's work, will defeat the inference from access and similarity. Keeping those records is the single most valuable thing a creative business can do to prepare for a claim it has not yet received.

What does not help the defendant is good faith. Copying need not be intentional. In Bright Tunes Music Corp. v. Harrisongs Music, Ltd., 420 F. Supp. 177 (S.D.N.Y. 1976), the court found that George Harrison had subconsciously copied the Chiffons' "He's So Fine" when he wrote "My Sweet Lord," accepted that he did not do it deliberately, and held him liable anyway; the Second Circuit affirmed the liability finding in ABKCO Music, Inc. v. Harrisongs Music, Ltd., 722 F.2d 988 (2d Cir. 1983). Three Boys applied the same doctrine to Michael Bolton. Intent affects willfulness and statutory damages, not liability.

Element two, part two: unlawful appropriation

What gets filtered out before anyone compares anything

Once copying in fact is established (or assumed, as it often is on a motion), the court must decide whether the defendant took enough protectable expression. The first move is subtraction. Every test, whatever its label, removes certain categories from the comparison because copyright does not protect them.

Ideas, systems, and methods. Baker v. Selden, 101 U.S. 99 (1879) held that a copyright in a book explaining a bookkeeping system did not give its author the exclusive right to the system or to the ruled forms needed to use it. Section 102(b) codified the principle. The line between an idea and its expression is not fixed, and Judge Learned Hand's "abstractions" passage in Nichols v. Universal Pictures Corp., 45 F.2d 119 (2d Cir. 1930) remains the standard description of the problem: as one strips away more and more of the incident from a play, there is a series of increasingly general patterns, and somewhere along that series the patterns stop being expression and become ideas, which are free. Nichols held that a play and a film sharing the premise of a Jewish family and an Irish family whose children secretly marry shared only an idea and stock characters. Hand admitted that "nobody has ever been able to fix that boundary, and nobody ever can."

Facts. Under Feist, facts are discovered rather than created and are never protectable, however much it cost to find them. Historical interpretation gets the same treatment: in Hoehling v. Universal City Studios, Inc., 618 F.2d 972 (2d Cir. 1980), a filmmaker who adopted an author's theory of who sabotaged the Hindenburg took an idea, not expression.

Merger. When an idea can be expressed in only one way or a handful of ways, the expression merges with the idea and is unprotectable, because protecting the expression would protect the idea. The Ninth Circuit's jeweled bee pin case, Herbert Rosenthal Jewelry Corp. v. Kalpakian, 446 F.2d 738 (9th Cir. 1971), is the classic statement; a bee-shaped pin covered in jewels can only look so many ways. Courts disagree about whether merger is a question of copyrightability or of infringement, and the difference matters for who bears the burden, but in practice it is argued at the filtration stage.

Scènes à faire. Elements that are standard, stock, or practically indispensable to a given topic are "scenes that must be done" and are filtered out. A police drama in the South Bronx will have drunks, prostitutes, and abandoned cars; a cop-buddy film will have a mismatched pair; a medieval fantasy will have a castle and a dragon; a database program will have a login screen. Each is free for the taking. The doctrine extends to functional constraints in software (compatibility requirements, hardware limits, industry conventions) and to the "standard" features of a product category.

Thin copyright. Works that barely clear the originality bar get protection only against virtually identical copying. The Ninth Circuit's clearest statement is Mattel, Inc. v. MGA Entertainment, Inc., 616 F.3d 904 (9th Cir. 2010), the Bratz case. Chief Judge Kozinski explained the sorting rule: where the range of possible expression is wide ("gazillions of ways to make an aliens-attack movie"), protection is broad and a substantially similar work infringes; where the range is narrow, protection is thin and only a virtually identical work does. Applying it, the court held that Carter Bryant's preliminary doll sculpt deserved only thin protection, because a fashion doll's idealized proportions can be exaggerated only so far before the doll looks freakish, while his sketches of complete dolls deserved broad protection, because face paint, hair, and clothing can vary endlessly. Even for the sketches, the district court had failed to filter out unprotectable ideas such as a bratty look or trendy clothing, and the copyright injunction was vacated. The same court's decision in Satava v. Lowry, 323 F.3d 805 (9th Cir. 2003), involving lifelike glass-in-glass jellyfish sculptures, held that the artist could not claim the natural form of a jellyfish, the clear glass shroud, or the standard proportions of the medium. His own contributions (the curl of particular tendrils, the arrangement of certain hues, the shape of the bells) earned only a thin copyright against virtually identical copying, so the preliminary injunction he had won against a competing glass artist was reversed.

Material the plaintiff did not create. Anything copied from the public domain or from third parties is subtracted under § 103(b), but the plaintiff's own choices about that material are not. In Tufenkian Import/Export Ventures, Inc. v. Einstein Moomjy, Inc., 338 F.3d 127 (2d Cir. 2003), a rug designer had cropped and elongated part of a public domain antique carpet and deleted many of its motifs to create a more open design. A competitor's rug reproduced those selections and deletions almost exactly, and the Second Circuit held that it infringed even though every individual motif was free for anyone to use.

What is left after subtraction is the plaintiff's original expression, and it is only that residue that the second step compares. The debate across circuits is about how to perform the comparison and who performs it.

The Second Circuit: the ordinary observer and the more discerning observer

The Second Circuit's test descends from Arnstein through Peter Pan Fabrics, Inc. v. Martin Weiner Corp., 274 F.2d 487 (2d Cir. 1960), where Judge Hand asked whether "the ordinary observer, unless he set out to detect the disparities, would be disposed to overlook them, and regard their aesthetic appeal as the same." The fabric designs in that case were close enough that an ordinary purchaser would not notice the differences, and that was infringement. The ordinary observer test is a holistic, impressionistic comparison; it does not dissect, it does not require expert help, and it is meant to be applied by a jury.

The problem with a pure ordinary observer test is that an ordinary observer does not know which elements are protectable. Two quilts that share a public domain alphabet layout look alike for reasons copyright does not care about. The Second Circuit therefore adopted a "more discerning ordinary observer" test for works that incorporate unprotectable elements. In Boisson v. Banian, Ltd., 273 F.3d 262 (2d Cir. 2001), involving alphabet quilts, the court held that the trial judge must first identify the unprotectable elements (the alphabet itself, the idea of a grid of letters) and then ask whether the works are substantially similar in the remaining protectable expression, including the original selection, arrangement, and combination of colors, shapes, and layout. Critically, the court warned against going too far: the more discerning observer must still compare "total concept and feel" and must not dissect the works into components so thoroughly that no original work could ever be infringed, because the arrangement of unprotectable elements can itself be protected.

Knitwaves, Inc. v. Lollytogs Ltd., 71 F.3d 996 (2d Cir. 1995), applied the test to children's sweaters with fall-leaf and squirrel motifs, and found infringement in the combination of the particular motifs, color palette, and layout even though a squirrel and a leaf are free to all. Tufenkian refined the test from both directions. The court agreed that public domain elements must be factored out, then reversed a judgment of non-infringement because the district court had let the defendant's few additions, and a different overall "feel," outweigh near-exact copying of the plaintiff's original choices about which public domain motifs to keep and which to drop. As Judge Learned Hand put it in Sheldon v. Metro-Goldwyn Pictures Corp., 81 F.2d 49 (2d Cir. 1936), "no plagiarist can excuse the wrong by showing how much of his work he did not pirate."

Two more Second Circuit decisions show the test in use at the dismissal stage. Peter F. Gaito Architecture, LLC v. Simone Development Corp., 602 F.3d 57 (2d Cir. 2010), held that a district court may decide substantial similarity on a Rule 12(b)(6) motion when the works themselves are before the court and no reasonable jury could find them substantially similar; the shared elements in two mixed-use development proposals were generalized concepts and site constraints. Zalewski v. Cicero Builder Dev., Inc., 754 F.3d 95 (2d Cir. 2014), affirmed summary judgment for builders accused of copying colonial house plans, explaining that the colonial style, the standard room layout, building codes, and market expectations were all unprotectable and that the plaintiff's copyright in his particular plans was thin.

The First, Third, Fifth, and Seventh Circuits apply versions of the ordinary observer test, so the Second Circuit's refinements travel; see, for example, Johnson v. Gordon, 409 F.3d 12 (1st Cir. 2005) (the "ordinary listener" in a music case), Tanksley v. Daniels, 902 F.3d 165 (3d Cir. 2018) (affirming dismissal of a claim that the series "Empire" copied a television pilot), Peel & Co. v. Rug Market, 238 F.3d 391 (5th Cir. 2001), and Atari, Inc. v. North American Philips Consumer Electronics Corp., 672 F.2d 607 (7th Cir. 1982) (the PAC-MAN case). The Fourth Circuit adds a twist: in Dawson v. Hinshaw Music Inc., 905 F.2d 731 (4th Cir. 1990), it held that the relevant observer is a member of the work's intended audience, so that choral arrangements are compared through the ears of choral directors, and in Lyons Partnership, L.P. v. Morris Costumes, Inc., 243 F.3d 789 (4th Cir. 2001) it held that a trial court erred by judging a Barney look-alike costume through the eyes of the adults who rented it rather than the young children who were its real audience. The Fourth Circuit otherwise uses a two-part extrinsic and intrinsic structure much like the Ninth Circuit's, judging the intrinsic step from the intended audience's perspective, and in Copeland v. Bieber, 789 F.3d 484 (4th Cir. 2015) it vacated a dismissal after listening to the songs for itself and concluding that a jury could hear the choruses as similar. In most cases the intended audience is the general public and the test collapses into the ordinary observer.

The Ninth Circuit: the extrinsic and intrinsic tests

The Ninth Circuit's test began in Sid & Marty Krofft Television Productions, Inc. v. McDonald's Corp., 562 F.2d 1157 (9th Cir. 1977), the "H.R. Pufnstuf" case, where McDonaldland's characters and setting were found to have copied the children's program. The court split the inquiry into an "extrinsic" test for similarity of ideas, which was objective and could use expert testimony and analytic dissection, and an "intrinsic" test for similarity of expression, which was subjective and left to the ordinary reasonable person without experts.

The labels have shifted. Since Shaw v. Lindheim, 919 F.2d 1353 (9th Cir. 1990), the extrinsic test is an objective comparison of specific expressive elements rather than ideas: for literary works, plot, themes, dialogue, mood, setting, pace, characters, and sequence of events; for music, melody, harmony, rhythm, pitch, tempo, phrasing, structure, chord progressions, and lyrics; for visual works, subject, pose, lighting, angle, and composition. Cavalier v. Random House, Inc., 297 F.3d 815 (9th Cir. 2002), made explicit that the extrinsic test compares only protectable elements, so filtration happens here. The intrinsic test then asks whether an ordinary reasonable person would find the "total concept and feel" of the works substantially similar, considering only the protectable elements that survived.

Three procedural consequences follow and they drive Ninth Circuit practice. First, a plaintiff must satisfy both tests, so failing the extrinsic test ends the case: Kouf v. Walt Disney Pictures & Television, 16 F.3d 1042 (9th Cir. 1994), affirmed summary judgment against a screenwriter who claimed "Honey, I Shrunk the Kids" copied his treatment, because the shared elements were general ideas. Second, because the extrinsic test is objective, the court can resolve it on a motion to dismiss or for summary judgment when the works are before it: Rentmeester v. Nike, Inc., 883 F.3d 1111 (9th Cir. 2018), affirmed dismissal of a photographer's claim that Nike's "Jumpman" logo and the photograph behind it copied his 1984 image of Michael Jordan in a grand jeté, because Rentmeester could not monopolize the idea of Jordan in a leaping, grand jeté-inspired pose, and Nike's photographer had not copied the details that made Rentmeester's expression of that pose original (the position of each limb, the setting, the lighting, the angle). The Ninth Circuit has also cautioned that dismissal is the exception. In Hanagami v. Epic Games, Inc., 85 F.4th 931 (9th Cir. 2023), it reversed the dismissal of a choreographer's claim against the maker of Fortnite, holding that a court may not break a dance into individual "poses," which are unprotectable alone, while ignoring the selection and arrangement of movement, timing, and transitions, and that a copied passage is not unprotectable merely because it is short. Third, the intrinsic test is for the jury alone, so once a plaintiff survives the extrinsic test, summary judgment for the defendant is nearly impossible, and expert testimony is confined to the extrinsic step. Funky Films, Inc. v. Time Warner Entertainment Co., 462 F.3d 1072 (9th Cir. 2006), applied the structure to "Six Feet Under" and a screenplay about a family-run funeral home, finding that beyond the premise the works diverged in plot, characters, and tone.

The Ninth Circuit has also accepted the "selection and arrangement" theory on which a combination of unprotectable elements can be protected if the combination is original and the defendant copied it. Skidmore confirmed the theory exists, but it also warned that a plaintiff must present it concretely, identifying which elements were combined and how, and that simply pointing to a list of commonplace features does not get a selection-and-arrangement instruction. Gray v. Hudson, 28 F.4th 87 (9th Cir. 2022), discussed below, shows what happens when that warning is ignored.

Software: abstraction, filtration, comparison

Computer programs are literary works under § 101 and get the same protection, but a program is also a functional machine, and the idea/expression and merger problems are acute. The Third Circuit's early approach in Whelan Associates, Inc. v. Jaslow Dental Laboratory, Inc., 797 F.2d 1222 (3d Cir. 1986), treated a program's purpose as its only idea and everything else, including its overall "structure, sequence and organization," as protectable expression. That approach gave software authors a very broad copyright and was widely criticized.

The Second Circuit replaced it in Computer Associates International, Inc. v. Altai, Inc., 982 F.2d 693 (2d Cir. 1992), which announced the three-step abstraction-filtration-comparison test. Abstraction breaks the program down into levels of generality, from its ultimate function down through architecture, modules, data structures, algorithms, and finally literal code, borrowing Nichols. Filtration removes at each level the elements dictated by efficiency (where there is one efficient way to do it, merger applies), dictated by external factors (hardware specs, compatibility with other software, industry standards, user demands), and taken from the public domain. Comparison then asks whether the defendant copied a substantial part of what remains, judged by its importance to the plaintiff's program. In Altai itself, a rewritten program that had been scrubbed of literally copied code shared only elements dictated by the operating system interfaces it had to call, and there was no infringement.

The Tenth Circuit adopted and elaborated the test in Gates Rubber Co. v. Bando Chemical Industries, Ltd., 9 F.3d 823 (10th Cir. 1993), listing six levels of abstraction (main purpose, program structure, modules, algorithms and data structures, source code, object code) and insisting that the court state its filtration findings on the record. The First Circuit went further on the protectability side in Lotus Development Corp. v. Borland International, Inc., 49 F.3d 807 (1st Cir. 1995), holding that the Lotus 1-2-3 menu command hierarchy was an uncopyrightable "method of operation" under § 102(b), so Borland's copying of it to make its spreadsheet compatible was not infringement; the Supreme Court affirmed by an equally divided Court in Lotus Development Corp. v. Borland International, Inc., 516 U.S. 233 (1996), which left the question open nationally.

The Ninth Circuit's software decisions sit on the fair use side as often as the similarity side. Sega Enterprises Ltd. v. Accolade, Inc., 977 F.2d 1510 (9th Cir. 1992), and Sony Computer Entertainment, Inc. v. Connectix Corp., 203 F.3d 596 (9th Cir. 2000), held that intermediate copying during reverse engineering to reach unprotected functional elements can be fair use, and the Federal Circuit, applying Ninth Circuit law in Atari Games Corp. v. Nintendo of America Inc., 975 F.2d 832 (Fed. Cir. 1992), accepted the same principle but ruled against Atari, which had obtained a copy of Nintendo's source code from the Copyright Office by falsely representing that litigation was pending and had copied more than reverse engineering required. The Second Circuit's decision in Softel, Inc. v. Dragon Medical & Scientific Communications, Inc., 118 F.3d 955 (2d Cir. 1997) is the essential corrective to over-filtration: a district court that filtered out each design element individually erred by failing to consider whether the combination of elements was protectable.

The largest software case of the century, Oracle America, Inc. v. Google LLC, produced two Federal Circuit decisions, Oracle America, Inc. v. Google Inc., 750 F.3d 1339 (Fed. Cir. 2014) (holding the declaring code and structure of the Java API packages copyrightable), and Oracle America, Inc. v. Google LLC, 886 F.3d 1179 (Fed. Cir. 2018) (reversing a jury's fair use verdict), before the Supreme Court decided Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021), on fair use alone. The Court assumed without deciding that the declaring code was copyrightable and held Google's copying of roughly 11,500 lines to reimplement 37 API packages in Android was fair use as a matter of law. The copyrightability question that Altai and Lotus framed therefore remains open at the Supreme Court level, and in the Ninth Circuit a defendant who reimplements an interface will argue Sega, § 102(b), and Google together. The fair use side is covered in Copyright Fair Use: The Four Factors, Transformative Use, and What Warhol Changed; software protection generally is in Legal Protection of Software: Copyrights, Patents, Trade Secrets, and Contracts.

The practical upshot for a software case: expect the court to require a filtration analysis, usually from an expert, that walks through the accused elements level by level and explains for each why it is or is not dictated by function, efficiency, compatibility, or convention. The side that produces the more disciplined filtration usually wins the motion. We provide a template in the Copyright Infringement Litigation Toolkit: Similarity Charts, Expert Reports, and Filtration Analyses.

The leading cases in music, literature, film, and art

The tests above are abstractions until you watch them decide real disputes. The cases below are the ones lawyers cite, judges quote, and clients have usually heard of, grouped by the kind of work involved. Read them for the reasoning, not the scorecard: the same court can find infringement in one fabric case and none in the next, and the difference is almost always in what survived filtration.

Music: the cases that taught everyone the rules

Music cases are where the doctrine gets tested most publicly, partly because juries can hear the works and partly because the building blocks of popular music (a handful of chords, common scales, standard song forms) are so limited that filtration decides nearly everything.

Three Boys Music (2000). The Isley Brothers' "Love Is a Wonderful Thing" and Michael Bolton's song of the same title shared a title phrase, a hook, and several melodic and rhythmic features. The jury found infringement and the Ninth Circuit affirmed, holding that the combination of five otherwise unprotectable elements could be protectable, that the access evidence (modest airplay during Bolton's youth, plus his documented familiarity with the Isleys) was enough, and that subconscious copying is actionable. Plaintiffs cite it for everything; defendants note that it predates Skidmore and that the court itself called the access evidence attenuated.

Swirsky v. Carey (2004). Swirsky v. Carey, 376 F.3d 841 (9th Cir. 2004), reversed summary judgment for Mariah Carey in a dispute over the chorus of "Thank God I Found You." The district court had compared the melodies measure by measure and found too few identical notes. The Ninth Circuit held that the extrinsic test cannot be reduced to a note-by-note count, that a musicologist's analysis of chord progression, rhythm, structure, and pitch sequence together was enough to raise a triable issue, and that a chorus could be protectable even if individual elements were common. It is the case that made musicologists indispensable.

Williams v. Gaye (2018). Williams v. Gaye, 895 F.3d 1106 (9th Cir. 2018), affirmed a jury verdict that "Blurred Lines" infringed Marvin Gaye's "Got to Give It Up," over a dissent arguing that the two songs shared only a style or groove and that the majority had allowed a jury to protect a genre. The majority emphasized the deferential standard of review of a verdict supported by competing expert testimony; it did not hold that a groove is protectable. Because the 1909 Act applied, the district court limited the Gaye composition to the deposit copy (six handwritten pages of sheet music), which kept the recording's cowbell, party noise, and much of its percussion and vocal performance out of the case; the Ninth Circuit accepted that ruling without deciding it, and Skidmore later settled the point. The decision alarmed songwriters, and courts have since read it narrowly.

Skidmore (2020). The en banc court's rejection of the inverse ratio rule and its treatment of the deposit copy, discussed above, reset Ninth Circuit music litigation. The opinion also affirmed the jury instruction that common musical elements (descending chromatic scales, arpeggios, short sequences of three notes) are not protectable in isolation.

Gray v. Hudson (2022). Gray v. Hudson involved the Christian hip-hop song "Joyful Noise" and Katy Perry's "Dark Horse." A jury found infringement based on an eight-note ostinato (a repeated instrumental figure) and awarded $2.8 million; the district court granted judgment as a matter of law and the Ninth Circuit affirmed. The ostinato used a scale fragment, even rhythm, and a timbre that were each commonplace, and the plaintiffs had not shown that the combination was original in a way that distinguished it from thousands of prior works. Commonplace elements like these, the court explained (quoting Skidmore), are "building blocks" that "belong in the public domain and cannot be exclusively appropriated by any particular author," and an unoriginal combination of them is no more protectable.

Hall v. Swift (2019). Hall v. Swift (9th Cir. 2019) (unpublished), reversed the dismissal of a claim that Taylor Swift's "Shake It Off" copied the lyric "playas, they gonna play, and haters, they gonna hate" from a 2001 song, because the complaint plausibly alleged that the lyric was original. The panel later amended its memorandum to delete broader language describing originality as normally a question of fact, but the reversal stood. The case later settled. It is the counterweight to Rentmeester: short phrases are usually unprotectable, but the Ninth Circuit will not always decide that at the Rule 12 stage.

Structured Asset Sales v. Sheeran (2024). In the Second Circuit, Structured Asset Sales, LLC v. Sheeran, 120 F.4th 1066 (2d Cir. 2024), affirmed summary judgment for Ed Sheeran on a claim that "Thinking Out Loud" copied Marvin Gaye's "Let's Get It On." The court held that the combination of a four-chord progression and a syncopated harmonic rhythm, each commonplace, was too unoriginal to be protected even as a combination, and it joined the Ninth Circuit in treating the deposit copy as the measure of a 1909 Act composition. A jury had earlier found for Sheeran in a parallel suit by the Townsend heirs.

Sampling. A sound recording copyright is separate from the composition, and sampling cases turn on the recording. The Sixth Circuit's Bridgeport Music, Inc. v. Dimension Films, 410 F.3d 792 (6th Cir. 2005), held that any unlicensed sampling of a sound recording infringes, however short, under its reading of § 114(b): "Get a license or do not sample." The Ninth Circuit expressly rejected that rule in VMG Salsoul, LLC v. Ciccone, 824 F.3d 871 (9th Cir. 2016), holding that a 0.23-second horn hit sampled into Madonna's "Vogue" was de minimis and that the ordinary de minimis rule applies to recordings. Outside the Sixth Circuit, VMG Salsoul is the majority view. For compositions, Newton v. Diamond, 388 F.3d 1189 (9th Cir. 2004), held that the Beastie Boys' licensed sample of a six-second, three-note flute passage took so little of the underlying composition that it was de minimis, and that performance elements unique to the recording are excluded from the composition comparison.

Literary, dramatic, and visual works

For stories, courts compare plot, sequence of events, characters, theme, mood, setting, pace, and dialogue, and the cases turn on how far up the abstraction ladder the similarities sit. Nichols is the baseline. Hoehling, discussed above, held that a nonfiction author's theory of the Hindenburg disaster, the historical facts, and the stock scenes of 1930s Germany were all free for a filmmaker to use. Walker v. Time Life Films, Inc., 784 F.2d 44 (2d Cir. 1986), held that "Fort Apache, The Bronx" did not infringe a police memoir because the shared elements (a precinct in a decayed neighborhood, drunks, prostitutes, rats, the murder of officers) were scènes à faire of the genre. Williams v. Crichton, 84 F.3d 581 (2d Cir. 1996), applied the same reasoning to "Jurassic Park" and children's dinosaur books. In the Ninth Circuit, Cavalier found no infringement in the stories as a whole but reversed summary judgment as to two pieces of artwork (a moon-shaped night light built into a board book's back cover and an illustration of stars lounging on clouds), a reminder that a defendant who borrows one protectable component can infringe even when the works as a whole differ. Benay v. Warner Bros. Entertainment, Inc., 607 F.3d 620 (9th Cir. 2010), found "The Last Samurai" shared only a premise with the plaintiffs' screenplay, though it sent an implied-contract claim for the idea itself back for trial, which is the usual fate of idea-submission disputes.

Characters can be protected independently if sufficiently delineated, which is why the Second Circuit in Warner Bros. Inc. v. American Broadcasting Cos., 720 F.2d 231 (2d Cir. 1983) compared Superman to "The Greatest American Hero" at the level of character and found a parody-tinged, bumbling hero was not substantially similar to the original. The court also held that a survey reporting that many viewers were reminded of Superman could not create a jury question when the works were not substantially similar as a matter of law, while leaving open whether a survey might help a jury in a genuinely close case. In the Ninth Circuit, a character is protected apart from any particular story only if it has physical as well as conceptual qualities, is sufficiently delineated to be recognizable whenever it appears, and is especially distinctive; the Batmobile passed that test in DC Comics v. Towle, 802 F.3d 1012 (9th Cir. 2015), and a set of emotion-themed children's characters failed it in Daniels v. Walt Disney Co., 958 F.3d 767 (9th Cir. 2020).

Visual works force courts to articulate what a photographer or illustrator actually authored. Mannion v. Coors Brewing Co., 377 F. Supp. 2d 444 (S.D.N.Y. 2005), is the most careful judicial treatment, separating originality in rendition (angle, lighting, lens, film, exposure), timing (being there when the moment happened), and creation of the subject (staging the scene). Steinberg v. Columbia Pictures Industries, Inc., 663 F. Supp. 706 (S.D.N.Y. 1987), found that the poster for "Moscow on the Hudson" infringed Saul Steinberg's famous New Yorker cover, "View of the World from 9th Avenue," because the poster copied the specific rendering style, perspective, and whimsical details, not merely the idea of a parochial view of the world. Leigh v. Warner Bros., Inc., 212 F.3d 1210 (11th Cir. 2000), involving the "Bird Girl" statue photograph from "Midnight in the Garden of Good and Evil," held that the statue, the cemetery, and the idea of photographing them were unprotectable, but the photographer's choices of light, angle, and mood could support a claim against similar film shots. Rentmeester and Ets-Hokin v. Skyy Spirits, Inc., 323 F.3d 763 (9th Cir. 2003), the vodka bottle case, confirm that a photograph of a subject does not give the photographer rights over the subject: once the bottle itself and the conventions of commercial product photography were filtered out under the merger and scènes à faire doctrines, the photographer was left with a thin copyright that a competitor's photographs of the same bottle, not virtually identical to his, did not infringe.

Useful articles present a threshold problem. Under Star Athletica, L.L.C. v. Varsity Brands, Inc., 580 U.S. 405 (2017), a design feature of a useful article (there, chevrons and stripes on cheerleading uniforms) is protectable only if it can be perceived as a work of art separate from the article and would qualify as a protectable pictorial, graphic, or sculptural work on its own. Only separable features enter the comparison.

De minimis copying

Even verbatim copying is not infringement if the amount taken is so trivial that the law does not notice it. In the Second Circuit the test is whether the copying is quantitatively and qualitatively insignificant and whether the ordinary observer would recognize the appropriation. Ringgold v. Black Entertainment Television, Inc., 126 F.3d 70 (2d Cir. 1997), held that a poster of a story quilt visible for a cumulative 27 seconds in a sitcom was not de minimis because it was clearly recognizable and used for its decorative purpose. Sandoval v. New Line Cinema Corp., 147 F.3d 215 (2d Cir. 1998), went the other way for photographs that appeared out of focus, in the background, and for seconds in "Seven." VMG Salsoul and Newton apply the doctrine to music. The doctrine concerns how much was taken, not how much harm resulted: in Bell v. Wilmott Storage Services, LLC, 12 F.4th 1065 (9th Cir. 2021), the Ninth Circuit held that a business that displayed an entire photograph on its website could not call the infringement de minimis on the theory that almost no one saw it. In practice the defense succeeds for fleeting, background, or unrecognizable uses and fails for anything a viewer would notice as the plaintiff's work.

The circuit tests compared

The differences are real but should not be overstated. Every circuit filters unprotectable material before comparing; every circuit asks some version of whether an ordinary audience would recognize the copying; every circuit lets the court end a case where the only similarities are ideas. What differs is the sequencing, the vocabulary, and the scope of expert testimony, and those differences matter most at summary judgment.

Experts, surveys, and the limits of lay opinion

Expert testimony in these cases is governed by Fed. R. Evid. 702 and the gatekeeping standard of Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), as extended to all expertise by Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999). The 2023 amendment to Rule 702, which clarified that the proponent must show admissibility by a preponderance and that an expert's opinion must reflect a reliable application of the method, has given defendants a sharper tool against musicologists and software experts who announce similarity without a reproducible method. See Expert Witnesses After the 2023 Amendment to Rule 702: Daubert, Disclosure, and Cross-Examination.

What an expert may testify about depends on the step. On copying in fact, experts are welcome everywhere: a musicologist can identify shared melodic contours and explain why they are unlikely to be coincidental; a software expert can find shared dead code or identical variable names; a forensic examiner can establish dates. On unlawful appropriation, the ordinary observer circuits traditionally exclude expert opinion on the ultimate question of similarity because the test is the lay reaction, while the Ninth Circuit admits it on the extrinsic test and excludes it on the intrinsic test. Software is the universal exception: Altai recognized that lay jurors cannot perform abstraction and filtration without help, and Gates Rubber treats expert filtration as expected. Music is the second exception, as Swirsky and Williams show. The best experts in all genres do three things: identify the similarities objectively, place them against a catalogue of prior art to show whether they are commonplace, and explain their significance within the plaintiff's work.

Surveys of consumer reaction, a staple of trademark litigation, rarely appear in copyright cases. Warner Bros. v. ABC explains why: substantial similarity "is not a concept familiar to the public at large," so a survey reporting that viewers were reminded of Superman could not override the court's ruling that the works were not substantially similar as a matter of law. The court expressly left open whether a well-designed survey could help a jury in a genuinely close case, and that door remains open but rarely used. Surveys occasionally appear on damages (apportioning profits to the infringing element) and in the Fourth Circuit's intended-audience inquiry. Lay opinion that "they sound the same" is generally excluded as unhelpful, though fact testimony about what a witness observed (a radio listener who mistook one song for the other) is sometimes admitted on copying in fact.

Procedure, timing, and cost: from the complaint to the verdict

The doctrine above is only half of how these cases are won. The other half is procedural: which questions the judge resolves early, which go to a jury, what evidence has to be in the record when it matters, and how much it costs to get there. A typical federal similarity case that does not settle runs eighteen months to three years from complaint to verdict, and most of the decisive events happen long before trial.

Pleading and the motion to dismiss

A complaint must state a plausible claim under Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007) and Ashcroft v. Iqbal, 556 U.S. 662 (2009), which in a copyright case means facts, not labels, on each element: the registration, the plaintiff's ownership, a concrete theory of access, and the specific protectable expression that was copied. "Upon information and belief, defendants had access to plaintiff's work" is the sentence that loses motions. Woodland shows the problem in the internet era: a plaintiff who alleges only that his work was online, without facts making it reasonably possible that the defendant saw it, has not pleaded access.

Similarity can be decided on the pleadings too. Under Fed. R. Civ. P. 10(c), exhibits attached to a complaint are part of it, and courts also consider works the complaint incorporates by reference. Gaito holds that in copyright cases "the works themselves supersede and control contrary descriptions of them," including the plaintiff's own characterizations in the complaint, so a court may compare the works on a Fed. R. Civ. P. 12(b)(6) motion and dismiss if no reasonable jury could find them substantially similar. The Second Circuit applied that rule in Abdin v. CBS Broadcasting Inc., 971 F.3d 57 (2d Cir. 2020), affirming dismissal of a claim that "Star Trek: Discovery" copied a video game because both used a space-traveling tardigrade, and the Seventh Circuit did the same in Hobbs v. John, 722 F.3d 1089 (7th Cir. 2013), where the songwriter of "Natasha" had even sent his song to the publisher that later released Elton John's "Nikita," but the two songs shared only the unprotectable idea of a doomed Cold War romance. The Ninth Circuit permits the same disposition (Rentmeester) but has reversed dismissals where the comparison turned on elements a judge could not responsibly assess without fact development or expert help, as in Hanagami. For a defendant, the practical rule is to move to dismiss only when the works speak for themselves and the shared elements are obviously ideas or stock material. For a plaintiff, attach the works, plead the protectable elements with precision, and plead the access chain with names and dates. Our Motions to Dismiss Under Rule 12: A Practical Guide for Plaintiffs and Defendants covers the general standards.

Discovery: what each side actually needs

Discovery in a similarity case is narrower than in most commercial litigation but deeper in a few places. The plaintiff needs the access trail: who received the submission, what the recipient did with it, whether it was forwarded, and who worked on the accused work. That usually means document requests under Fed. R. Civ. P. 34, depositions of the creators and their collaborators, and subpoenas under Fed. R. Civ. P. 45 to labels, publishers, agencies, studios, and platforms whose submission logs or analytics show who opened what. Office routines matter: in Bouchat v. Baltimore Ravens, Inc., 241 F.3d 350 (4th Cir. 2001), the jury was entitled to rely on testimony about how faxes sent to the stadium authority's chairman were routinely handled, in an office suite the Ravens' owner shared, to find that the team's logo designers had access to an amateur artist's sketch. By contrast, Jones v. Blige, 558 F.3d 485 (6th Cir. 2009) and Loomis v. Cornish, 836 F.3d 991 (9th Cir. 2016) affirmed summary judgment where the plaintiff could show only that the work reached someone at a large company, or reached a town where the defendants once spent ten days, without a link to the people who wrote the accused song.

The defendant needs the plaintiff's creative history (to find the plaintiff's own sources, which may also be the defendant's) and must preserve and produce its own: drafts, sketches, session files, version control history, and the metadata that dates them. That creative history is both the independent creation defense and, if it is missing, the plaintiff's best spoliation argument; see Litigation Holds, Spoliation, and Rule 37(e): Preserving Electronic Evidence Before It Costs You the Case. Software cases add source code review under a protective order with strict inspection protocols, which Protective Orders, Confidentiality Designations, and Sealing: Secrecy and Its Limits in Federal Court explains.

Two discovery failures recur in the reported cases. First, the works themselves must be in the record, in the version that matters. In Seiler v. Lucasfilm, Ltd., 808 F.2d 1316 (9th Cir. 1987), the best evidence rule barred a plaintiff from proving infringement with drawings he reconstructed after the fact, and in Antonick v. Electronic Arts, Inc., 841 F.3d 1062 (9th Cir. 2016), a jury verdict for a programmer was erased because neither side's source code was ever introduced, so the jury had nothing to compare; the court cited Airframe Systems, Inc. v. L-3 Communications Corp., 658 F.3d 100 (1st Cir. 2011), where the plaintiff could not produce the registered version of its code. Second, requests for admission under Fed. R. Civ. P. 36 have teeth: in Jones v. Blige, plaintiffs who failed to answer requests that the lyrics were not substantially similar and that two defendants had no access were deemed to have admitted both points.

Summary judgment

Under Fed. R. Civ. P. 56, Anderson v. Liberty Lobby, Inc., 477 U.S. 242 (1986), and Celotex Corp. v. Catrett, 477 U.S. 317 (1986), summary judgment turns on whether a reasonable jury could find for the nonmovant, and similarity cases are decided this way more often than the "question of fact" label suggests. In the Second Circuit, a court may grant summary judgment for the defendant when the similarity concerns only unprotectable elements or when no reasonable observer could find the works substantially similar (Zalewski). In the Ninth Circuit, failing the extrinsic test ends the case, while passing it sends the case to a jury on the intrinsic test, so the summary judgment motion is really a fight about filtration (Funky Films; L.A. Printex Industries, Inc. v. Aeropostale, Inc., 676 F.3d 841 (9th Cir. 2012), which reversed summary judgment for the defendant because a jury could find both access through the sale of 50,000 yards of fabric and extrinsic similarity). Plaintiffs can win summary judgment too: in Unicolors, Inc. v. Urban Outfitters, Inc., 853 F.3d 980 (9th Cir. 2017), the Ninth Circuit affirmed summary judgment of infringement where two fabric designs were close enough that no reasonable juror could find otherwise, leaving only willfulness and damages for trial.

What a court may not do at this stage is choose between credible experts. That was the error in Repp, and it is why a plaintiff with a qualified expert who has done a real filtration and prior art analysis usually survives in the circuits that let experts speak to similarity. In software cases, some courts now front-load the protectability fight. In SAS Institute, Inc. v. World Programming Ltd., 64 F.4th 1319 (Fed. Cir. 2023), applying Fifth Circuit law, the Federal Circuit approved a pretrial "copyrightability hearing" and held that once the defendant presented persuasive evidence that the asserted elements of a statistical software system were unprotectable, the copyright owner had to identify with specificity which elements it claimed and show that they were protected, following the burden-shifting framework of Compulife Software Inc. v. Newman, 959 F.3d 1288 (11th Cir. 2020). See Summary Judgment Under Rule 56: Winning and Surviving Before Trial.

Jury instructions, trial, and post-trial motions

Most similarity trials are decided by the instructions. Under Fed. R. Civ. P. 51, a party must request an instruction and object to the court's instructions on the record, or later review is limited to plain error, which is exactly what doomed the plaintiff's selection-and-arrangement argument in Skidmore. A sound set of instructions defines ownership and the registered work (for 1909 Act works, the deposit copy), explains that copying may be proved by access plus probative similarity or by striking similarity, tells the jury that independent creation defeats the claim, lists the categories of unprotectable material, states the circuit's test for unlawful appropriation, and, if the plaintiff relies on a combination theory, explains that the combination must itself be original. The Ninth Circuit's Manual of Model Civil Jury Instructions, chapter 17, is a common starting point even outside that circuit, and a special verdict form under Fed. R. Civ. P. 49 that asks separately about access, copying, and protectable expression makes the verdict reviewable.

Presentation is its own discipline. The jury should hear or see the works as the law defines them, which in a 1909 Act music case means a performance of the deposit copy rather than the famous recording (Skidmore; Williams v. Gaye). Demonstratives that transpose keys, isolate notes, or overlay images are effective and dangerous: the opposing expert should test every one, and Fed. R. Evid. 403 and Fed. R. Evid. 1006 govern what comes in. After the verdict, Fed. R. Civ. P. 50 and Fed. R. Civ. P. 59 give the losing party a second chance, and similarity verdicts are reversed on that basis more often than most verdicts: Selle (judgment notwithstanding the verdict for the Bee Gees), Gray (judgment as a matter of law after a $2.8 million verdict), and Antonick all ended that way. A party that wants to renew a Rule 50 motion after the verdict must have raised the same ground before the case went to the jury. For the mechanics, see Post-Trial Motions in Federal Court: Judgment as a Matter of Law, New Trial, Remittitur, and Relief From Judgment.

What these cases cost and how they end

Budgets vary enormously, but the pattern is predictable. In our experience, a dispute that ends on a motion to dismiss costs each side somewhere between the high five figures and the low six figures. A case litigated through fact discovery, dueling expert reports, and summary judgment commonly costs each side several hundred thousand dollars and can pass a million, and a music or software trial with competing experts can cost each side several million. Expert work is a large share of that: musicologists, software forensic experts, and damages experts bill by the hour, and a full report, rebuttal report, deposition, and trial testimony is a six-figure exercise in a serious case. The defendant's creative-process discovery is often more expensive than anyone expects, because it means collecting years of drafts, session files, or repositories from individual creators.

The fee-shifting rules shape the economics as much as the merits do. Under 17 U.S.C. § 505, the court may award a reasonable attorney's fee to the prevailing party, plaintiff or defendant, on an even-handed basis: Fogerty v. Fantasy, Inc., 510 U.S. 517 (1994) arose when John Fogerty was sued for sounding too much like his own earlier Creedence Clearwater Revival song, won at trial, and was held entitled to the same consideration for fees a winning plaintiff would get. Kirtsaeng v. John Wiley & Sons, Inc., 579 U.S. 197 (2016) then directed courts to give substantial weight to the objective reasonableness of the losing party's position while considering all the circumstances. Because § 412 bars fees and statutory damages for many plaintiffs who registered late, a defendant can face a plaintiff who cannot recover fees while the defendant can, and that asymmetry drives settlement. "Full costs" under § 505 means only the categories in 28 U.S.C. § 1920, not expert fees or e-discovery vendors, after Rimini Street, Inc. v. Oracle USA, Inc., 586 U.S. 334 (2019). A defendant can also use an offer of judgment under Fed. R. Civ. P. 68; because Marek v. Chesny, 473 U.S. 1 (1985) holds that Rule 68 "costs" include attorney's fees when the underlying statute defines fees as costs, and § 505 does, the Eleventh Circuit in Jordan v. Time, Inc., 111 F.3d 102 (11th Cir. 1997) shifted a defendant's post-offer fees to a plaintiff who recovered less than the offer, while the Seventh Circuit in Harbor Motor Co. v. Arnell Chevrolet-Geo, Inc., 265 F.3d 638 (7th Cir. 2001) held that only a prevailing party can recover § 505 fees, Rule 68 or not. Check your circuit before you rely on either rule.

On the damages side, 17 U.S.C. § 504(b) lets the plaintiff recover actual damages plus the infringer's profits not already counted, with a burden-shifting rule that favors plaintiffs: the plaintiff proves the infringer's gross revenue, and the infringer must prove its deductible expenses and the share of profit attributable to something other than the copied work. Apportionment can cut an award sharply, as it did in Sheldon v. Metro-Goldwyn Pictures Corp., 309 U.S. 390 (1940), where the Court approved awarding the playwrights one fifth of a film's profits; in music cases, actual damages are often built on a hypothetical license, and in Williams v. Gaye the Ninth Circuit affirmed actual damages calculated as a 50 percent share of publishing revenue, profits awards against two defendants, and a running royalty of half the future songwriter and publishing revenue from "Blurred Lines." Statutory damages under § 504(c) are the alternative for timely registered works. Damages may reach back as far as the claim is timely: Warner Chappell Music, Inc. v. Nealy, 601 U.S. 366 (2024) held that a copyright owner with a timely claim under the discovery rule may recover damages for infringement that occurred more than three years before suit. Injunctions under 17 U.S.C. § 502 are discretionary, and after eBay Inc. v. MercExchange, L.L.C., 547 U.S. 388 (2006) and Salinger v. Colting, 607 F.3d 68 (2d Cir. 2010) there is no presumption of irreparable harm.

Most similarity disputes end without a verdict. Music claims usually settle with a songwriting credit and a share of the publishing income; software and design claims settle with a license, a redesign, or a withdrawal from the market; literary and film claims, which plaintiffs lose at a high rate, often end on a motion followed by a fee application. For claims worth $30,000 or less, the Copyright Claims Board offers a cheaper, voluntary forum that respondents can opt out of, covered in The Copyright Claims Board: Small Claims Copyright Litigation Under the CASE Act. For the settlement mechanics, including Rule 68, see Settlement Agreements, Releases, and Rule 68: Ending a Case on Terms That Hold.

Strategy for plaintiffs and defendants

For the plaintiff, or the person thinking about suing

1. Register first, and register the right version. No suit without a registration decision (Fourth Estate), no statutory damages or fees without timely registration (§ 412), and no comparison without the registered version in the record (Airframe; Antonick). If the work was unpublished, register the draft that existed before the defendant's work appeared. Special handling at the Copyright Office, for an added fee, shortens the wait from months to days.
2. Build the access chain before you file. Identify every person who received the work, every platform where it appeared, and every intermediary between those people and the defendant's creators. Dates matter: access after the accused work was created is not access.
3. Do your own filtration before the defendant does it for you. Make a list of every similarity, then honestly mark which ones are ideas, facts, genre conventions, public domain material, or elements you borrowed. If what remains is thin, your claim is thin, and the case value and strategy should reflect that.
4. Hire the expert early and ask for a prior-art search. A musicologist or software expert who has already searched for earlier works sharing the same features is worth far more than one who discovers them at deposition.
5. Plead a selection-and-arrangement theory deliberately, if you have one. Identify the elements, explain how their particular combination is original, and request the instruction. Skidmore and Gray are what happens otherwise.
6. Price the fee risk. A weak claim against a well-funded defendant can end with a fee award against you under Kirtsaeng.

For the defendant, or the person who just received a demand letter

1. Preserve and collect your creation record immediately. Drafts, session files, version control history, sketches, emails with collaborators, and the dates on all of them are your independent creation defense. Send a litigation hold the day the letter arrives.
2. Map your own sources. If your work and the plaintiff's share a common ancestor (a public domain melody, a genre convention, a stock photo, an open source library), that is the most effective answer to an inference of copying.
3. Test the plaintiff's ownership and registration. Ask who the author is, whether a transfer was signed, whether the registration covers the version in suit, and whether anything in the application was knowingly inaccurate under § 411(b) and Unicolors.
4. Choose the battlefield. Where the works are short and the overlap is obviously unprotectable, a motion to dismiss with the works attached can end the case in months. Where it is not, build the summary judgment record around filtration and prior art.
5. Consider the declaratory judgment option and the forum. A party facing threats may sue first under 28 U.S.C. § 2201 when there is a real controversy; see Declaratory Judgment Actions: Jurisdiction, Discretion, and the Race to the Courthouse.
6. Use the fee statute both ways. A strong defense, documented early, supports a § 505 motion at the end and a Rule 68 offer in the middle. For the first response to a letter, see Responding to a Cease and Desist Letter: A Practical Guide.

Common mistakes

• Confusing the two kinds of similarity. Plaintiffs prove copying with unprotectable similarities and then forget that the second step excludes them; defendants argue that unprotectable similarities prove nothing at the first step, which is wrong.
• Treating "it's online" as access. After Woodland, Art Attacks, and Loomis, a plaintiff needs more than availability on the internet or presence in the same city.
• Counting similarities instead of weighing them. Twenty trivial overlaps are worth less than one distinctive, protectable passage copied nearly verbatim, and courts say so.
• Comparing the wrong version. The recording instead of the deposit copy, the current build instead of the registered release, the finished film instead of the screenplay that was actually submitted.
• Letting the expert skip prior art. Selle and Gray were both lost after jury verdicts for the plaintiffs, because the plaintiffs' experts never dealt with earlier works and commonplace material, a gap a prior-art search would have exposed at the start.
• Losing the works. No source code in evidence, no originals, no deposit copy: Antonick and Seiler are the result.
• Ignoring requests for admission. Unanswered requests become admissions (Jones v. Blige).
• Forgetting to request instructions and object. Rule 51 forfeiture decided an important part of Skidmore.
• Overlooking registration timing. A plaintiff who discovers at the end of the case that § 412 forecloses statutory damages and fees has usually overinvested in the case.

A worked example: the demo, the producer, and the hit

Suppose Mara Quinlan, a songwriter in Nashville, records a demo of her song "Paper Lanterns" in October 2022 and, after meeting an A&R coordinator for Lodestar Records at a showcase, emails him a private streaming link. In mid-2024, Lodestar releases "Lantern Light," performed by Kade Rivers and co-written by producer Toby Wren and two others. It is a hit. Its chorus opens with a six-note hook that matches Mara's in pitch and rhythm, lands on the same unusual melisma on the second syllable, and sits over the same four-chord progression; one verse also uses the phrase "paper lanterns." Mara is certain she was copied. Here is how the case actually unfolds.

Ownership and registration. Mara never registered the demo. She files an application in August 2024 with special handling and receives a registration within days, so she can sue (Fourth Estate). But because the work was unpublished and the alleged infringement began before registration, § 412 bars statutory damages and attorney's fees. Her recovery is limited to actual damages and Lodestar's profits attributable to the infringement, and she faces the risk of paying Lodestar's fees if she loses badly. That single fact changes the economics of the whole case.

Access. The email to the A&R coordinator is a start, but Jorgensen and Jones v. Blige say that receipt by someone at a large company is not enough without a link to the writers. Mara's lawyers subpoena the streaming service and learn that the private link was played four times in November 2022 from an IP address at Lodestar's Los Angeles office. Document requests then turn up the label's internal "pitch folder," whose access log shows that Wren's assistant downloaded "Paper Lanterns" in December 2022. Now there is a chain of events of the Bouchat kind, and access is a jury question at worst.

Independent creation and common sources. Wren produces his session files. They show the hook developing over three drafts, but the earliest is dated February 2023, after the download, so the files help Mara more than Wren. Wren's musicologist then does what the experts in Selle and Gray did not do: a prior-art search. She finds the same six pitches in a 1970s gospel standard and in two earlier pop songs, though with a different rhythm and over different chords.

Filtration. The phrase "paper lanterns" is a short phrase and unprotectable on its own under the Copyright Office's regulation, 37 C.F.R. § 202.1. The four-chord progression and the syncopated rhythm are commonplace, as Sheeran holds for a similar combination, and the pitch sequence alone is in the prior art. What survives is the particular combination of that pitch sequence, that rhythm, that melisma, and that placement at the top of the chorus over that progression, a selection-and-arrangement theory that Mara's lawyers must articulate concretely and carry all the way into a requested jury instruction (Skidmore).

The test and the forum. Mara could sue in Tennessee or in California if each court has personal jurisdiction, and venue lies under 28 U.S.C. § 1400(a) where a defendant resides or may be found. In the Sixth Circuit, Kohus requires the court to filter first and then ask whether an ordinary listener would find the protectable elements substantially similar. In the Ninth Circuit, the hook must survive the extrinsic test, after which the intrinsic question belongs to the jury. Lodestar, headquartered in Los Angeles, would rather be in California, where Gray and Skidmore are binding and judges are experienced at filtering commonplace musical material; Mara might prefer home.

Summary judgment and value. With strong access evidence, a combination that does not appear in the prior art, and a qualified expert on each side, summary judgment for Lodestar is unlikely in either circuit, but so is a large verdict, because apportionment will separate the value of the hook from Kade Rivers' performance, the production, and the marketing. The realistic resolution is a settlement that gives Mara a co-writer credit and a share of the publishing income, the outcome many music cases reach once the access evidence is in. Change one fact (Mara's email went only to a general submissions inbox and nothing connects it to Wren) and the case looks like Jorgensen or Loomis: Mara would then need striking similarity, and the gospel standard in the prior art would make that showing very hard.

The lessons travel beyond music. Registration timing set the remedies, the access chain was built from logs rather than intuition, the defense's best evidence came from its own files and a prior-art search, and the case value turned on filtration and apportionment rather than on how alike the two choruses sounded.

Frequently asked questions

Does the defendant have to copy on purpose? No. Subconscious copying is still copying, as George Harrison learned in Bright Tunes and Michael Bolton in Three Boys. Intent matters to willfulness and the size of statutory damages, not to liability.

How much can someone copy before it counts as infringement? There is no safe number of notes, words, seconds, or percentage points, and the "change it by 30 percent" rule is a myth. Courts look at both the quantity and the quality of the protectable expression taken; a short passage that is the heart of the work can be enough, and a large amount of unprotectable material is not. As Judge Hand put it, "no plagiarist can excuse the wrong by showing how much of his work he did not pirate."

Can anyone own a style, a genre, or a vibe? No, but the line between a style and its expression is where the hard cases live. Williams v. Gaye shows the risk of getting close to a specific work's protectable expression while borrowing its feel; Gray and Skidmore show the limits on protecting commonplace building blocks. See The Fine Line: Copyright Protection for Style, Typefaces, and Fonts.

Is posting my work online enough to prove that someone had access to it? Usually not. Woodland rejected access based on Instagram posts with a few dozen likes, and Art Attacks rejected it for designs sold at county fairs and shown on a slow-loading website. Wide dissemination means real reach, and a chain of events means evidence connecting the work to the people who made the accused work.

Can a judge throw out my case without a jury? Yes. When the works are before the court and the only similarities are unprotectable, courts dismiss on the pleadings (Gaito, Rentmeester, Abdin) or grant summary judgment (Zalewski, Kouf). In the Ninth Circuit, though, a plaintiff who survives the extrinsic test is entitled to have a jury apply the intrinsic test.

Do I need an expert? In music and software cases, almost always, because the court needs help separating protectable from commonplace material, and the side without a credible expert usually loses the filtration fight. In literary and visual cases, the works often speak for themselves, though an expert on genre conventions or photographic technique can still help. Whatever the medium, put the works themselves in evidence; an expert cannot substitute for them (Antonick).

What if the work was made with AI? Copyright protects only human authorship, so purely machine-generated material cannot be the basis of a claim, and AI-assisted works are protected only to the extent of their human contributions; see Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025) and Copyright in AI-Assisted Works: Human Authorship, Thaler v. Perlmutter, and the Copyright Office Guidance. Claims that AI models infringe by training on or reproducing works raise different questions, discussed in Copyright Infringement Claims Against Generative AI: The New York Times, Getty, and What Comes Next.

How long do I have to sue? Three years from when the claim accrues under § 507(b), with each infringing act starting its own clock (Petrella). Most circuits apply a discovery rule to accrual, and under Nealy a timely claim can recover damages for older infringements. Waiting is still costly, because evidence of access and independent creation disappears.

Practical takeaways

• Keep the two similarity questions separate: probative similarity (with access) proves copying; substantial similarity of protectable expression proves unlawful appropriation.
• Access must be reasonably possible, not merely conceivable. Build it from documents, logs, and witnesses, and remember that internet availability alone rarely suffices.
• Filtration decides most cases. Ideas, facts, scènes à faire, merger, public domain material, and thin copyrights shrink the comparison before anyone compares anything.
• Know your circuit's test, who applies it, and when it can end the case: the Second Circuit's more discerning observer, the Ninth Circuit's extrinsic and intrinsic tests, the Sixth and D.C. Circuits' filter-then-compare approach, and abstraction-filtration-comparison for software.
• Put the right version of each work in the record, from the registration through trial.
• Experts win or lose these cases on method: objective identification of similarities, a real prior-art search, and an honest account of significance.
• Registration timing under § 412 and fee shifting under § 505 often matter more to the outcome than the merits, so price them in at the start.
• Creative businesses should keep dated drafts and source records as a matter of routine; they are the cheapest insurance against a claim not yet received.

This document is general information, not legal advice for any particular situation; the law changes, and the facts matter.

Related Documents

Related Articles

• Copyright Damages and Remedies: Statutory Damages, Profits, Injunctions, and Fees: what a winning plaintiff recovers and how registration timing limits it.
• Copyright Fair Use: The Four Factors, Transformative Use, and What Warhol Changed: the main defense when copying is admitted.
• Copyright Ownership, Joint Authorship, and Termination of Transfers: proving the first element and the chain of title.
• Copyright Registration: A Comprehensive Guide: the registration prerequisite and how to satisfy it quickly.
• Legal Protection of Software: Copyrights, Patents, Trade Secrets, and Contracts: where copyright fits among the other protections for code.
• Summary Judgment Under Rule 56: Winning and Surviving Before Trial: the general standards that frame similarity motions.
• Expert Witnesses After the 2023 Amendment to Rule 702: Daubert, Disclosure, and Cross-Examination: the gatekeeping rules for musicologists and software experts.
• Expert Depositions: Taking Apart the Opinion, Protecting Your Own Expert, and Building the Daubert Record: how to test a similarity expert before trial.
• The Copyright Claims Board: Small Claims Copyright Litigation Under the CASE Act: the low-cost alternative for small claims.
• Post-Trial Motions in Federal Court: Judgment as a Matter of Law, New Trial, Remittitur, and Relief From Judgment: how verdicts like those in Selle and Gray were undone.

Related Guides

• Litigating Substantial Similarity in a Copyright Case: A Practical Guide: the step-by-step companion to this article, from pre-suit investigation to verdict.
• Motions to Dismiss Under Rule 12: A Practical Guide for Plaintiffs and Defendants: pleading standards and early motions.
• Responding to a Cease and Desist Letter: A Practical Guide: the first moves after an accusation arrives.
• Taking and Defending an Expert Deposition: A Practical Guide: preparing and cross-examining the experts these cases depend on.

Related Checklists

• Copyright Infringement Claim Evaluation Checklist: A Practical Checklist: a step-by-step screen of any claim, for either side.
• Work Made for Hire Determination Checklist: A Practical Checklist: who owns a work created by an employee or contractor.
• Litigation Hold and Evidence Preservation Checklist: A Practical Checklist: preserving drafts, session files, and repositories.

Related Toolkits

• Copyright Infringement Litigation Toolkit: Similarity Charts, Expert Reports, and Filtration Analyses: templates for the comparison charts, filtration memos, and expert reports discussed here.
• Fair Use Toolkit: Analysis Memoranda, Clearance Decisions, and Litigation Positions: forms for the defense that often travels with a similarity case.
• Expert Witness Toolkit: retention, disclosure, and challenge materials for expert-driven cases.