Teva Pharma. USA, Inc. v. Sandoz, Inc., 574 U.S. 318 (2015)
Teva’s patent covers a multiple sclerosis drug. When Sandoz tried to market a generic version of the drug, Teva sued for infringement. Sandoz countered that the patent was invalid because a claim that the active ingredient had “a molecular weight of 5 to 9 kilodaltons” was indefinite under 35 U. S. C. 112, for not stating which of three methods was used to determine that weight. The district court upheld the patent. Reversing, the Federal Circuit reviewed all aspects of claim construction de novo, including the determination of subsidiary facts. The Supreme Court vacated. When reviewing a district court’s resolution of subsidiary factual matters made during construction of a patent claim, the Federal Circuit must apply a “clear error,” not a de novo, standard of review. FRCP 52(a)(6) states: a court of appeals must not set aside “[f]indings of fact” unless they are “clearly erroneous.” Clear error review is particularly important in patent cases because a district judge has more opportunity to gain “familiarity with specific scientific problems and principles” than an appeals judge who must read a written transcript. When reviewing only evidence intrinsic to the patent, the judge’s determination is a determination of law, and the court of appeals will review that construction de novo; where the court needs to consult disputed extrinsic evidence to understand, for example, background science, courts need to make subsidiary factual findings about the extrinsic evidence. The ultimate construction of the claim is a legal conclusion subject to de novo review, but to overturn resolution of an underlying factual dispute, the appellate court must find that the judge, in respect to those findings, committed clear error. Here, the district court made a factual finding, crediting Teva’s expert’s account about how a skilled artisan would understand molecular weights. When the Federal Circuit reviewed the decision, it failed to accept that explanation without finding that the determination was “clearly erroneous.”
De novo review is appropriate for an appeal based on intrinsic elements of a patent application, which require a court to make determinations of law. However, a clear error standard of review is appropriate for an appeal based on extrinsic elements of a patent application, since these require only determinations of fact.
NOTE: Where it is feasible, a syllabus (headnote) will be released, as is being done in connection with this case, at the time the opinion is issued. The syllabus constitutes no part of the opinion of the Court but has been prepared by the Reporter of Decisions for the convenience of the reader. See United States v. Detroit Timber & Lumber Co., 200 U.S. 321, 337.
SUPREME COURT OF THE UNITED STATES
teva pharmaceuticals usa, inc., et al. v. sandoz, inc., et al.
certiorari to the united states court of appeals for the federal circuit
No. 13–854. Argued October 15, 2014—Decided January 20, 2015
Petitioners, Teva Pharmaceuticals (and related firms), own a patent that covers a manufacturing method for the multiple sclerosis drug Copaxone. When respondents, Sandoz, Inc. (and other firms), tried to market a generic version of the drug, Teva sued them for patent infringement. Sandoz countered that the patent was invalid. Specifically, Sandoz argued that the claim that Copaxone’s active ingredient had “a molecular weight of 5 to 9 kilodaltons” was fatally indefinite, see 35 U. S. C. §112 ¶2, because it did not state which of three methods of calculation—the weight of the most prevalent molecule, the weight as calculated by the average weight of all molecules, or weight as calculated by an average in which heavier molecules count for more—was used to determine that weight. After considering conflicting expert evidence, the District Court concluded that the patent claim was sufficiently definite and the patent was thus valid. As relevant here, it found that in context a skilled artisan would understand that the term “molecular weight” referred to molecular weight as calculated by the first method. In finding the “molecular weight” term indefinite and the patent invalid on appeal, the Federal Circuit reviewed de novo all aspects of the District Court’s claim construction, including the District Court’s determination of subsidiary facts.
Held: When reviewing a district court’s resolution of subsidiary factual matters made in the course of its construction of a patent claim, the Federal Circuit must apply a “clear error,” not a de novo, standard of review. Pp. 4–16.
(a) Federal Rule of Civil Procedure 52(a)(6) states that a court of appeals “must not . . . set aside” a district court’s “[f]indings of fact” unless they are “clearly erroneous.” It sets out a “clear command,” Anderson v. Bessemer City, 470 U.S. 564, 574, and “does not make exceptions or . . . exclude certain categories of factual findings” from the court of appeals’ obligation, Pullman-Standard v. Swint, 456 U.S. 273, 287. The Rule thus applies to both subsidiary and ultimate facts. Ibid. And the function of an appeals court reviewing the findings of a “ ‘district court sitting without a jury . . . is not to decide factual issues de novo.’ ” Anderson, supra, at 573. Even if exceptions to the Rule were permissible, there is no convincing ground for creating an exception here. Markman v. Westview Instruments, Inc., 517 U.S. 370, neither created, nor argued for, such an exception. There, the Court held that the ultimate question of claim construction is for the judge, not the jury, id., at 372, but it did not thereby create an exception from the ordinary rule governing appellate review of factual matters. Instead, the Court pointed out that a judge, in construing a patent claim, is engaged in much the same task as the judge would be in construing other written instruments, such as deeds, contracts, or tariffs. Id., at 384, 386, 388, 389. Construction of written instruments often presents a “question solely of law,” at least when the words in those instruments are “used in their ordinary meaning.” Great Northern R. Co. v. Merchants Elevator Co., 259 U.S. 285, 291. But if a written instrument uses “technical words or phrases not commonly understood,” id., at 292, those words may give rise to a factual dispute. If so, extrinsic evidence may help to “establish a usage of trade or locality.” Ibid. And in that circumstance, the “determination of the matter of fact” will “preced[e]” the “function of construction.” Ibid.
The Markman Court also recognized that courts will sometimes have to resolve subsidiary factual disputes in patent construction; Rule 52 requires appellate courts to review such disputes under the “clearly erroneous” standard. Application of this standard is further supported by precedent and by practical considerations. Clear error review is “particularly” important in patent cases because a district court judge who has presided over, and listened to, the entire proceeding has a comparatively greater opportunity to gain the necessary “familiarity with specific scientific problems and principles,” Graver Tank & Mfg. Co. v. Linde Air Products Co., 339 U.S. 605, 610, than an appeals court judge who must read a written transcript or perhaps just those portions referenced by the parties. Pp. 4–8.
(b) Arguments to the contrary are unavailing. Sandoz claims that separating “factual” from “legal” questions may be difficult and, like the Federal Circuit, posits that it is simpler for the appellate court to review the entirety of the district court’s claim construction de novo than to apply two separate standards. But courts of appeals have long been able to separate factual from legal matters, see, e.g., First Options of Chicago, Inc. v. Kaplan, 514 U.S. 938, 947–948, and the Federal Circuit’s efforts to treat factual findings and legal conclusions similarly have brought with them their own complexities. As for Sandoz’s argument that “clear error” review will bring about less uniformity, neither the Circuit nor Sandoz has shown that divergent claim construction stemming from divergent findings of fact on subsidiary matters should occur more than occasionally. Pp. 8–11.
(c) This leaves the question of how the clear error standard should be applied when reviewing subsidiary factfinding in patent claim construction. When the district court reviews only evidence intrinsic to the patent, the judge’s determination is solely a determination of law, and the court of appeals will review that construction de novo. However, where the district court needs to consult extrinsic evidence in order to understand, for example, the background science or the meaning of a term in the relevant art during the relevant time period, and where those subsidiary facts are in dispute, courts will need to make subsidiary factual findings about the extrinsic evidence. The district judge, after deciding the factual dispute, will then interpret the patent claim in light of the facts as he has found them. The ultimate construction of the claim is a legal conclusion that the appellate court can review de novo. But to overturn the judge’s resolution of an underlying factual dispute, the appellate court must find that the judge, in respect to those factual findings, has made a clear error. Pp. 11–14.
(d) Here, for example, the District Court made a factual finding, crediting Teva’s expert’s account, and thereby rejecting Sandoz’s expert’s contrary explanation, about how a skilled artisan would understand the way in which a curve created from chromatogram data reflects molecular weights. Based on that factual finding, the District Court reached the legal conclusion that figure 1 did not undermine Teva’s argument that molecular weight referred to the first method of calculating molecular weight. When the Federal Circuit reviewed the District Court’s decision, it did not accept Teva’s expert’s explanation, and it failed to accept that explanation without finding that the District Court’s contrary determination was “clearly erroneous.” The Federal Circuit erred in failing to review this factual finding only for clear error. Teva asserts that there are two additional instances in which the Federal Circuit rejected the District Court’s factual findings without concluding that they were clearly erroneous; those matters are left for the Federal Circuit to consider on remand. Pp. 14–16.
723 F.3d 1363, vacated and remanded.
Breyer, J., delivered the opinion of the Court, in which Roberts, C. J., and Scalia, Kennedy, Ginsburg, Sotomayor, and Kagan, JJ., joined. Thomas, J., filed a dissenting opinion, in which Alito, J., joined.
In Markman v. Westview Instruments, Inc.,517 U.Â S. 370 (1996), we explained that a patent claim is that â€œportion of the patent document that defines the scope of the patenteeâ€™s rights.â€ Id., at 372. We held that â€œthe construction of a patent, including terms of art within its claim,â€ is not for a jury but â€œexclusivelyâ€ for â€œthe courtâ€ to determine. Ibid. That is so even where the construction of a term of art has â€œevidentiary underpinnings.â€ Id., at 390. Todayâ€™s case involves claim construction with â€œevidentiary underpinnings.â€ See Part III, infra. And, it requires us to determine what standard the Court of Appeals should use when it reviews a trial judgeâ€™s resolution of an underlying factual dispute. Should the Court of Appeals review the district courtâ€™s factfinding de novo as it would review a question of law? Or, should it review that factfinding as it would review a trial judgeâ€™s factfinding in other cases, namely by taking them as correct â€œunless clearly erroneous?â€ See Fed. Rule Civ. Proc. 52(a)(6). We hold that the appellate court must apply a â€œclear error,â€ not a de novo, standard of review. I The basic dispute in this case concerns the meaning of the words â€œmolecular weightâ€ as those words appear in a patent claim. The petitioners, Teva Pharmaceuticals (along with related firms), own the relevant patent. The patent covers a manufacturing method for Copaxone, a drug used to treat multiple sclerosis. The drugâ€™s active ingredient, called â€œcopolymer-1,â€ is made up of molecules of varying sizes. App. 1143a. And the relevant claim describes that ingredient as having â€œa molecular weight of 5 to 9 kilodaltons.â€ Id., at 1145a. The respondents, Sandoz, Inc. (and several other firms), tried to market a generic version of Copaxone. Teva sued Sandoz for patent infringement. 810 F.Â Supp. 2d 578, 581 (SDNY 2011). Sandoz defended the suit by arguing that the patent was invalid. Ibid. The Patent Act requires that a claim â€œparticularly poin[t] out and distinctly clai[m] the subject matter which the applicant regards as his invention.â€35 U.Â S.Â C. Â§112 Â¶2 (2006 ed.); see Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.Â S. ___, ___, n. 1 (2014) (slip op., at 3, n. 1)). The phrase â€œmolecular weight of 5 to 9 kilodaltons,â€ said Sandoz, did not satisfy this requirement. The reason that the phrase is fatally indefinite, Sandoz argued, is that, in the context of this patent claim, the term â€œmolecular weightâ€ might mean any one of three different things. The phrase might refer (1) to molecular weight as calculated by the weight of the molecule that is most prevalent in the mix that makes up copolymer-1. (The scientific term for molecular weight so calculated is, we are told, â€œpeak average molecular weight.â€) The phrase might refer (2) to molecular weight as calculated by taking all the different-sized molecules in the mix that makes up copolymer-1 and calculating the average weight, i.e., adding up the weight of each molecule and dividing by the number of molecules. (The scientific term for molecular weight so calculated is, we are told, â€œnumber average molecular weight.â€) Or, the phrase might refer (3) to molecular weight as calculated by taking all the different-sized molecules in the mix that makes up copolymer-1 and calculating their average weight while giving heavier molecules a weight-related bonus when doing so. (The scientific term for molecular weight so calculated, we are told, is â€œweight average molecular weight.â€) See 723 F.Â 3d 1363, 1367 (CA Fed. 2013); App. 124a. In Sandozâ€™s view, since Tevaâ€™s patent claim does not say which method of calculation should be used, the claimâ€™s phrase â€œmolecular weightâ€ is indefinite, and the claim fails to satisfy the critical patent law requirement. The District Court, after taking evidence from experts, concluded that the patent claim was sufficiently definite. Among other things, it found that in context a skilled artisan would understand that the term â€œmolecular weightâ€ referred to molecular weight as calculated by the first method, i.e., â€œpeak average molecular weight.â€ 810 F.Â Supp. 2d, at 596; see Nautilus, supra, at ___ (slip op., at 12) (â€œ[T]he definiteness inquiry trains on the understanding of a skilled artisan at the time of the patent applicationâ€). In part for this reason, the District Court held the patent valid. 810 F.Â Supp. 2d, at 596. On appeal, the Federal Circuit held to the contrary. It found that the term â€œmolecular weightâ€ was indefinite. And it consequently held the patent invalid. 723 F.Â 3d, at 1369. In reaching this conclusion, the Federal Circuit reviewed de novo all aspects of the District Courtâ€™s claim construction, including the District Courtâ€™s determination of subsidiary facts. Id., at 1369, 1373; see also Light-ing Ballast Control LLC v. Philips Electronics NorthAm. Corp., 744 F. 3d 1272, 1276â€“1277 (CA Fed. 2014) (en banc) (reaffirming de novo review of district court claim construction). Teva filed a petition for certiorari. And we granted that petition. The Federal Circuit reviews the claim construction decisions of federal district courts throughout the Nation, and we consequently believe it important to clarify the standard of review that it must apply when doing so. II A Federal Rule of Civil Procedure 52(a)(6) states that a court of appeals â€œmust not .Â .Â . set asideâ€ a district courtâ€™s â€œ[fÂ ]indings of factâ€ unless they are â€œclearly erroneous.â€ In our view, this rule and the standard it sets forth must apply when a court of appeals reviews a district courtâ€™s resolution of subsidiary factual matters made in the course of its construction of a patent claim. We have made clear that the Rule sets forth a â€œclear command.â€ Anderson v. Bessemer City,470 U.Â S. 564,574 (1985). â€œIt does not make exceptions or purport to exclude certain categories of factual findings from the obligation of a court of appeals to accept a district courtâ€™s findings unless clearly erroneous.â€ Pullman-Standard v. Swint,456 U. S. 273,287 (1982). Accordingly, the Rule applies to both subsidiary and ultimate facts. Ibid. And we have said that, when reviewing the findings of a â€œÂ â€˜district court sitting without a jury, appellate courts must constantly have in mind that their function is not to decide factual issues deÂ novo.â€™Â â€ Anderson, supra, at 573 (quoting Zenith Radio Corp. v. Hazeltine Research, Inc.,395 U.Â S. 100,123 (1969)). Even if exceptions to the Rule were permissible, we cannot find any convincing ground for creating an exception to that Rule here. The Rules Advisory Committee pointed out that, in general, exceptions â€œwould tend to undermine the legitimacy of the district courts .Â .Â . , multiply appeals .Â .Â . , and needlessly reallocate judicial author-ity.â€ Advisory Committeeâ€™s 1985 Note on subd. (a) of Fed. Rule Civ. Proc. 52, 28 U.Â S.Â C. App., pp. 908â€“909; see also Anderson, supra, at 574â€“575 (de novo review of fac-tual findings â€œwould very likely contribute only negligiblyâ€ to accuracy â€œat a huge cost in diversion of judicialresourcesâ€). Our opinion in Markman neither created, nor argued for, an exception to Rule 52(a). The question presented in that case was a Seventh Amendment question: Should a jury or a judge construe patent claims? 517 U.Â S., at 372. We pointed out that history provides no clear answer. Id., at 388. The task primarily involves the construction of written instruments. Id., at 386, 388, 389. And that task is better matched to a judgeâ€™s skills. Id., at 388 (â€œThe construction of written instruments is one of those things that judges often do and are likely to do better than jurors unburdened by training in exegesisâ€). We consequently held that claim construction falls â€œexclusively within the province of the court,â€ not that of the jury. Id., at 372. When describing claim construction we concluded that it was proper to treat the ultimate question of the proper construction of the patent as a question of law in the way that we treat document construction as a question of law. Id., at 388â€“391. But this does not imply an exception to Rule 52(a) for underlying factual disputes. We used the term â€œquestion of lawâ€ while pointing out that a judge, in construing a patent claim, is engaged in much the same task as the judge would be in construing other written instruments, such as deeds, contracts, or tariffs. Id., at 384, 386, 388, 389; see also Motion Picture Patents Co. v. Universal Film Mfg. Co.,243 U.Â S. 502,510 (1917) (patent claims are â€œaptly likened to the description in a deed, which sets the bounds to the grant which it containsâ€); Goodyear Dental Vulcanite Co. v. Davis,102 U.Â S. 222,227 (1880) (analogizing patent construction to the construction of other written instruments like contracts). Construction of written instruments often presents a â€œquestion solely of law,â€ at least when the words in those instruments are â€œused in their ordinary meaning.â€ Great Northern R. Co. v. Merchants Elevator Co., 259 U. Â S. 285, 291 (1922). But sometimes, say when a written instrument uses â€œtechnical words or phrases not commonly understood,â€ id., at 292, those words may give rise to a factual dispute. If so, extrinsic evidence may help to â€œestablish a usage of trade or locality.â€ Ibid. And in that circumstance, the â€œdetermination of the matter of factâ€ will â€œpreced[e]â€ the â€œfunction of construction.â€ Ibid.; see also 12 R. Lord, Williston on Contracts Â§Â§34:1, p.Â 2, 34:19, p. 174 (4th ed. 2012) (In contract interpretation, the existence of a â€œusageâ€â€”a â€œpractice or methodâ€ in the relevant industryâ€”â€œis a question of factâ€ (internal quotation marks omitted)). This factual determination, like all other factual determinations, must be reviewed for clear error. See Pullman-Standard, supra at 287 (The Rule does not â€œexclude certain categories of factual findingsâ€ and applies to both â€œsubsidiaryâ€ and â€œultimateâ€ facts (internal quotation marks omitted)). Accordingly, when we held in Markman that the ultimate question of claim construction is for the judge and not the jury, we did not create an exception from the ordinary rule governing appellate review of factual matters. Markman no more creates an exception to Rule 52(a) than would a holding that judges, not juries, determine equit-able claims, such as requests for injunctions. A conclusion that an issue is for the judge does not indicate that Rule 52(a) is inapplicable. See Fed. Rule Civ. Proc. 52 (setting the standard of review for â€œ[Factual] Findings and Conclusions by the Courtâ€ (emphasis added)). While we held in Markman that the ultimate issue of the proper construction of a claim should be treated as a question of law, we also recognized that in patent construction, subsidiary factfinding is sometimes necessary. Indeed, we referred to claim construction as a practice with â€œevidentiary underpinnings,â€ a practice that â€œfalls somewhere between a pristine legal standard and a simple historical fact.â€ 517 U.Â S., at 378, 388, 390. We added that sometimes courts may have to make â€œcredibility judgmentsâ€ about witnesses. Id., at 389. In other words, we recognized that courts may have to resolve subsidiary factual disputes. And, as explained above, the Rule requires appellate courts to review all such subsidiary factual findings under the â€œclearly erroneousâ€ standard. Precedent further supports application of the â€œclearly erroneousâ€ standard. Before the creation of the Federal Circuit, the Second Circuit explained that in claim construction, the subsidiary â€œquestion .Â .Â . of how the art un-derstood the term .Â .Â . was plainly a question of fact; and unless the [district courtâ€™s] finding was â€˜clearly erroneous,â€™ we are to takeâ€ it â€œas controlling.â€ Harries v. Air King Products, Co., 183 F. 2d 158, 164 (CA2 1950) (L. Hand, C.Â J.). We have said the same as to subsidiary factual findings concerning other patent law inquiries, including â€œobviousness.â€ Dennison Mfg. Co. v. Panduit Corp.,475 U.Â S. 809,811 (1986) (per curiam) (â€œsubsidiary determinations of the District Courtâ€ subject to Rule 52(a)â€™s clear error standard). Finally, practical considerations favor clear error review. We have previously pointed out that clear error review is â€œparticularlyâ€ important where patent law is at issue because patent law is â€œa field where so much depends upon familiarity with specific scientific problems and principles not usually contained in the general storehouse of knowledge and experience.â€ Graver Tank & Mfg. Co. v. Linde Air Products Co.,339 U.Â S. 605,610 (1950). A district court judge who has presided over, and listened to, the entirety of a proceeding has a comparatively greater opportunity to gain that familiarity than an appeals court judge who must read a written transcript or perhaps just those portions to which the parties have referred. Cf. Lighting Ballast, 744 F. 3d, at 1311 (Oâ€™Malley, J., dissenting) (Federal Circuit judges â€œlack the tools that district courts have available to resolve factual disputes fairly and accurately,â€ such as questioning the experts, examining the invention in operation, or appointing a court-appointed expert); Anderson, 470 U.Â S., at 574 (â€œThe trial judgeâ€™s major role is the determination of fact, and with experience in fulfilling that role comes expertiseâ€). B Sandoz argues that claim construction mostly consists of construing a set of written documents that do not give rise to subsidiary factual disputes. Tr. of Oral Arg. 39. It adds that separating â€œfactualâ€ from â€œlegalâ€ questions is often difficult. And Sandoz, like the Federal Circuit itself, argues that it is simpler for that appellate court to review the entirety of the district courtâ€™s claim construction de novo rather than to apply two separate standards. Id., at 38; see also Lighting Ballast, supra, at 1284 (criticizing clear error review in part because of the purportedly difficult task of â€œdisentanglingâ€ fact from law). But even were we free to ignore the Federal Rule (which we are not), we would not find this argument convincing. Courts of appeals have long found it possible to separate factual from legal matters. See, e.g., First Options of Chicago, Inc. v. Kaplan,514 U.Â S. 938â€“948 (1995) (review of factual findings for clear error and legal conclusions deÂ novo is the â€œordinaryâ€ standard for courts of appeals). At the same time, the Federal Circuitâ€™s efforts to treat factual findings and legal conclusions similarly have brought with them their own complexities. See e.g., Cybor Corp. v. FAS Technologies, Inc., 138 F. 3d 1448, 1454 (CA Fed. 1998) (en banc) (claim construction does not involve â€œfactual evidentiary findingsâ€ (citation and internal quotation marks omitted)); Lighting Ballast, supra, at 1284 (claim construction has â€œarguably factual aspectsâ€); Dow Jones & Co. v. Ablaise Ltd., 606 F.Â 3d 1338, 1344â€“1345 (CA Fed. 2010) (â€œ[T]his court,â€ while reviewing claim construction without deference, â€œtakes into account the views of the trial judgeâ€); Nazomi Communications Inc., v. Arm Holdings, PLC, 403 F. 3d 1364, 1371 (CA Fed. 2005) (â€œ[C]ommon sense dictates that the trial judgeâ€™s view will carry weightâ€ (citation and internal quotation marks omitted)); Lightning Ballast, supra, at 1294 (Lourie, J., concurring) (we should â€œrarelyâ€ overturn district courtâ€™s true subsidiary factfinding; â€œwe should, and do, give proper informal deference to the work of judges of a subordinate tribunalâ€); Cybor, supra, at 1480 (opinion of Newman, J.) (â€œBy continuing the fiction that there are no facts to be found in claim interpretations, we confound rather than ease the litigation processâ€); see also Anderson, supra, at 575 (the parties â€œhave already been forced to concentrate their energies and resources on persuading the trial judge that their account of the facts is the correct one; requiring them to persuade three more judges at the appellate level is requiring too muchâ€); Brief for Peter S. Menell et al. as Amici Curiae 5 (Federal Circuit overturns district court claim construction at unusually high rate). Finally, the Circuit feared that â€œclear errorâ€ review would bring about less uniformity. Lighting Ballast, supra, at 1280. Neither the Circuit nor Sandoz, however, has shown that (or explained why) divergent claim construction stemming from divergent findings of fact (on subsidiary matters) should occur more than occasionally. After all, the Federal Circuit will continue to review de novo the district courtâ€™s ultimate interpretation of the patent claims. And the attorneys will no doubt bring cases construing the same claim to the attention of the trial judge; those prior cases will sometimes be binding because of issue preclusion, see Markman, 517 U.Â S., at 391, and sometimes will serve as persuasive authority. Moreover, it is always possible to consolidate for discovery different cases that involve construction of the same claims. And, as we said in Markman, subsidiary factfinding is unlikely to loom large in the universe of litigated claim construction. Id., at 389â€“390. C The dissent argues that claim construction does not involve any â€œfactfinding,â€ or, if it does, claim construction factfinding is akin to the factfinding that underlies our interpretation of statutes. Post, at 1, 5â€“7 (opinion of Thomas, J.). Its first, broader contention runs contrary to our recognition in Markman that claim construction has â€œevidentiary underpinningsâ€ and that courts construing patent claims must sometimes make â€œcredibility judgmentsâ€ about witnesses. 517 U.Â S., at 389â€“390. Indeed, as discussed in Part III, infra, this case provides a perfect example of the factfinding that sometimes underlies claim construction: The parties here presented the District Court with competing fact-related claims by different experts, and the District Court resolved the issues of fact that divided those experts. The dissentâ€™s contention also runs contrary to Sandozâ€™s concession at oral argument that claim construction will sometimes require subsidiary factfinding. Tr. of Oral Arg. 33â€“34, 38â€“40. It is in tension with our interpretation of related areas of patent law, such as the interpretation of â€œobviousness,â€ which we have said involves subsidiary factfinding subject to Rule 52(a)â€™s clear error review. See Dennison, 475 U. S., at 811. And it fights the question presented in this case, which assumes the existence of such factfinding. See Pet. for Cert. i (whether â€œa district courtâ€™s factual finding in support of its construction of a patent claim term may be reviewed deÂ novo, . . . or only for clear errorâ€). Neither do we find factfinding in this context sufficiently similar to the factfinding that underlies statutory interpretation. Statutes, in general, address themselves to the general public; patent claims concern a small portion of that public. Statutes typically (though not always) rest upon congressional consideration of general facts related to a reasonably broad set of social circumstances; patents typically (though not always) rest upon consideration by a few private parties, experts, and administrators of more narrowly circumscribed facts related to specific technical matters. The public, and often an adversarial public, typically considers and discusses the relevant general facts before Congress enacts a statute; only private parties, experts, and administrators likely consider the relevant technical facts before the award of a patent. Given these differences, it is not surprising that this Court has never previously compared patent claim construction in any here relevant way to statutory construction. As discussed supra, at 5, however, the Court has repeatedly compared patent claim construction to the construction of other written instruments such as deeds and contracts. See, e.g., Markman, supra, at 384, 386, 388, 389; Motion Picture Patent Co., 243 U.Â S., at 510; Goodyear, 102 U.Â S., at 227. D Now that we have set forth why the Federal Circuit must apply clear error review when reviewing subsidiary factfinding in patent claim construction, it is necessary to explain how the rule must be applied in that context. We recognize that a district courtâ€™s construction of a patent claim, like a district courtâ€™s interpretation of a written instrument, often requires the judge only to examine and to construe the documentâ€™s words without requiring the judge to resolve any underlying factual disputes. As all parties agree, when the district court reviews only evidence intrinsic to the patent (the patent claims and specifications, along with the patentâ€™s prosecution history), the judgeâ€™s determination will amount solely to a determination of law, and the Court of Appeals will review that construction de novo. See Brief for Petitioners 27, Re-ply Brief 16; Brief for Respondents 43; see also Brief for United States as Amicus Curiae 12â€“13. In some cases, however, the district court will need to look beyond the patentâ€™s intrinsic evidence and to consult extrinsic evidence in order to understand, for example, the background science or the meaning of a term in the relevant art during the relevant time period. See, e.g., Seymour v. Osborne, 11 Wall. 516, 546 (1871) (a patent may be â€œso interspersed with technical terms and terms of art that the testimony of scientific witnesses is indispensable to a correct understanding of its meaningâ€). In cases where those subsidiary facts are in dispute, courts will need to make subsidiary factual findings about that extrinsic evidence. These are the â€œevidentiary underpinningsâ€ of claim construction that we discussed in Markman, and this subsidiary factfinding must be reviewed for clear error on appeal. For example, if a district court resolves a dispute between experts and makes a factual finding that, in general, a certain term of art had a particular meaning to a person of ordinary skill in the art at the time of the invention, the district court must then conduct a legal analysis: whether a skilled artisan would ascribe that same meaning to that term in the context of the specific patent claim under review. That is because â€œ[e]xperts may be examined to explain terms of art, and the state of the art, at any given time,â€ but they cannot be used to prove â€œthe proper or legal construction of any instrument of writing.â€ Winans v. New York & Erie R. Co., 21 How. 88, 100â€“101 (1859); see also Markman, supra, at 388 (â€œÂ â€˜Where technical terms are used, or where the qualities of substances .Â .Â . or any similar data necessary to the comprehension of the language of the patent are unknown to the judge, the testimony of witnesses may be received upon these subjects, and any other means of information be employed. But in the actual interpretation of the patent the court proceeds upon its own responsibility, as an arbiter of the law, giving to the patent its true and final character and forceâ€™Â â€ (quoting 2 W. Robinson, Law of Patents Â§732, pp. 482â€“483 (1890); emphasis in original)). Accordingly, the question we have answered here concerns review of the district courtâ€™s resolution of a subsidiary factual dispute that helps that court determine the proper interpretation of the written patent claim. The district judge, after deciding the factual dispute, will then interpret the patent claim in light of the facts as he has found them. This ultimate interpretation is a legal conclusion. The appellate court can still review the district courtâ€™s ultimate construction of the claim de novo. But, to overturn the judgeâ€™s resolution of an underlying factual dispute, the Court of Appeals must find that the judge, in respect to those factual findings, has made a clear error. Fed. Rule Civ. Proc. 52(a)(6). In some instances, a factual finding will play only a small role in a judgeâ€™s ultimate legal conclusion about the meaning of the patent term. But in some instances, a factual finding may be close to dispositive of the ultimate legal question of the proper meaning of the term in the context of the patent. Nonetheless, the ultimate question of construction will remain a legal question. Simply because a factual finding may be nearly dispositive does not render the subsidiary question a legal one. â€œ[A]n issue does not lose its factual character merely because its resolution is dispositive of the ultimateâ€ legal question. Miller v. Fenton,474 U.Â S. 104,113 (1985). It is analogous to a judge (sitting without a jury) deciding whether a defendant gave a confession voluntarily. The answer to the legal question about the voluntariness of the confes-sion may turn upon the answer to a subsidiary factual question, say â€œwhether in fact the police engaged in the intimidation tactics alleged by the defendant.â€ Id., at 112. An appellate court will review the trial judgeâ€™s factual determination about the alleged intimidation deferentially (though, after reviewing the factual findings, it will review a judgeâ€™s ultimate determination of voluntariness de novo). See id., at 112â€“118. An appellate court similarly should review for clear error those factual findings that underlie a district courtâ€™s claim construction. III We can illustrate our holding by considering an instance in which Teva, with the support of the Solicitor General, argues that the Federal Circuit wrongly reviewed the District Courtâ€™s factual finding de novo. See Brief for Petitioners 54â€“56; Brief for United States as Amicus Curiae 31â€“32. Recall that Tevaâ€™s patent claim specifies an active ingredient with a â€œmolecular weight of about 5 to 9 kilodaltons.â€ Recall Sandozâ€™s basic argument, namely that the term â€œmolecular weightâ€ is indefinite or ambiguous. The term might refer to the weight of the most numerous molecule, it might refer to weight as calculated by the average weight of all molecules, or it might refer to weight as calculated by an average in which heavier molecules count for more. The claim, Sandoz argues, does not tell us which way we should calculate weight. See Part I, supra. To illustrate, imagine we have a sample of copolymer-1 (the active ingredient) made up of 10 molecules: 4 weigh 6 kilodaltons each, 3 weigh 8 kilodaltons each, and 3 weigh 9 kilodaltons each. Using the first method of calculation, the â€œmolecular weightâ€ would be 6 kilodaltons, the weight of the most prevalent molecule. Using the second method, the molecular weight would be 7.5 (total weight, 75, di-vided by the number of molecules, 10). Using the third method, the molecular weight would be more than 8, depend-ing upon how much extra weight we gave to the heavier molecules. Teva argued in the District Court that the term â€œmolecular weightâ€ in the patent meant molecular weight calculated in the first way (the weight of the most prevalent molecule, or peak average molecular weight). Sandoz, however, argued that figure 1 of the patent showed that Teva could not be right. 810 F.Â Supp. 2d, at 590. (We have set forth figure 1 in the Appendix, infra). That figure, said Sandoz, helped to show that the patent term did not refer to the first method of calculation. Figure 1 shows how the weights of a sampleâ€™s molecules were distributed in three different samples. The curves indicate the number of molecules of each weight that were present in each of the three. For example, the figureâ€™s legend says that the first sampleâ€™s â€œmolecular weightâ€ is 7.7. According to Teva, that should mean that molecules weighing 7.7 kilodaltons were the most prevalent molecules in the sample. But, look at the curve, said Sandoz. It shows that the most prevalent molecule weighed, not 7.7 kilodaltons, but slightly less than 7.7 (about 6.8) kilodaltons. See App. 138aâ€“139a. After all, the peak of the first molecular weight distribution curve (the solid curve in the figure) is not at precisely 7.7 kilodaltons, but at a point just before 7.7. Thus, argued Sandoz, the figure shows that the patent claim term â€œmolecular weightâ€ did not mean molecular weight calculated by the first method. It must mean something else. It is indefinite. 810 F.Â Supp. 2d, at 590. The District Court did not accept Sandozâ€™s argument. Tevaâ€™s expert testified that a skilled artisan would understand that converting data from a chromatogram to molecular weight distribution curves like those in figure 1 would cause the peak on each curve to shift slightly; this could explain the difference between the value indicated by the peak of the curve (about 6.8) and the value in the figureâ€™s legend (7.7). App. 138aâ€“139a. Sandozâ€™s expert testified that no such shift would occur. App. 375aâ€“376a. The District Court credited Tevaâ€™s expertâ€™s account, thereby rejecting Sandozâ€™s expertâ€™s explanation. 810 F.Â Supp. 2d, at 589; Brief for Respondents 61. The District Courtâ€™s finding about this matter was a factual findingâ€”about how a skilled artisan would understand the way in which a curve created from chromatogram data reflects molecular weights. Based on that factual finding, the District Court reached the legal conclusion that figure 1 did not undermine Tevaâ€™s argument that molecular weight referred to the first method of calculation (peak average molecular weight). 810 F.Â Supp. 2d, at 590â€“591. When the Federal Circuit reviewed the District Courtâ€™s decision, it recognized that the peak of the curve did not match the 7.7 kilodaltons listed in the legend of figure 1. 723 F.Â 3d, at 1369. But the Federal Circuit did not accept Tevaâ€™s expertâ€™s explanation as to how a skilled artisan would expect the peaks of the curves to shift. And it failed to accept that explanation without finding that the District Courtâ€™s contrary determination was â€œclearly erroneous.â€ See ibid. The Federal Circuit should have accepted the District Courtâ€™s finding unless it was â€œclearly erroneous.â€ Our holding today makes clear that, in failing to do so, the Federal Circuit was wrong. Teva claims that there are two additional instances in which the Federal Circuit rejected the District Courtâ€™s factual findings without concluding that they were clearly erroneous. We leave these matters for the Federal Circuit to consider on remand in light of todayâ€™s opinion. We vacate the Federal Circuitâ€™s judgment, and we remand the case for further proceedings consistent with this opinion. It is so ordered. APPENDIX FIG. 1 (with minor additions to emphasize that the peak of the solid curve does not correspond precisely to 7.7kDa)