Patents

When Claim Construction Is a Fact: Teva Pharmaceuticals v. Sandoz and the Clear-Error Standard

The Supreme Court held that subsidiary factual findings underlying a claim construction must be reviewed for clear error, narrowing decades of de novo appellate review.

September 15, 2025
Laboratory chromatography equipment with a scientist reviewing a printed graph
The disputed term 'molecular weight' turned on expert testimony about how a chromatogram is read—a factual finding the Supreme Court said deserves deference. Shutterstock
Educational content, not legal advice. This article explains general legal concepts. It does not create an attorney–client relationship. For your specific situation, consult a licensed attorney.

In Teva Pharmaceuticals USA, Inc. v. Sandoz, Inc., 574 U.S. 318 (2015) (No. 13-854), the Supreme Court returned to claim construction to answer a question that Markman and Phillips had left smoldering: when a district court resolves a disputed term partly by finding facts about how skilled artisans understand technical language, how should an appellate court review those findings? Decided January 20, 2015, in a 7-2 opinion by Justice Breyer, the Court held that subsidiary factual findings must be reviewed for “clear error”—not de novo—rejecting the Federal Circuit’s long-standing practice of reviewing every aspect of a construction afresh. The decision recalibrated the relationship between trial and appellate courts in patent cases without disturbing Markman’s core holding that the judge, not the jury, construes claims.

At a glance

  • Court and date: Supreme Court of the United States, decided January 20, 2015 (7-2; Justices Thomas and Alito dissenting).
  • Docket: No. 13-854 (on certiorari to the U.S. Court of Appeals for the Federal Circuit).
  • Technology: Teva’s Copaxone (glatiramer acetate), a multiple-sclerosis therapy; the dispute concerned patents claiming a copolymer of a specified “molecular weight.”
  • Disputed term: “molecular weight”—indefinite on its face because at least three measures exist (peak average, number average, and weight average), each yielding a different number.
  • District court finding: Crediting expert testimony, the court found that a skilled artisan reading the specification would understand “molecular weight” to mean peak average molecular weight, derived directly from the size-exclusion chromatography described in the patent.
  • Holding: The Federal Circuit must review such subsidiary factual findings for clear error under Federal Rule of Civil Procedure 52(a)(6); the ultimate construction remains a legal conclusion reviewed de novo.

The doctrinal knot

Markman held that claim construction is a question for the judge, and in Cybor Corp. v. FAS Technologies, Inc., 138 F.3d 1448 (Fed. Cir. 1998), the Federal Circuit drew what seemed the logical conclusion: if construction is a question of law, appellate review is de novo, top to bottom. That tidy rule had a cost. Construing a technical term frequently requires a judge to resolve genuinely factual disputes—what a person of ordinary skill in a particular field would have understood a word to mean, how a laboratory technique actually works, what an industry custom was at the time. Under Cybor, an appellate panel could substitute its own view of those matters for the trial judge’s, even though the trial judge had heard live expert testimony and the appellate court had not.

Teva surfaced that tension cleanly. The word “molecular weight” is ambiguous to a chemist: it can denote peak average (Mp), number average (Mn), or weight average (Mw), and the three diverge. Sandoz argued the claims were indefinite because they did not say which. The district court, after weighing competing experts, found that a skilled artisan would read the term as peak average, because Mp is what one obtains directly from the chromatogram and calibration curve produced by the analytical method the patent described. On de novo review, the Federal Circuit reached the opposite result and held the claims indefinite. The question for the Supreme Court was whether that fresh review was proper.

Rule 52(a)(6) and the limits of de novo review

The Court grounded its answer in a procedural rule of general application. Federal Rule of Civil Procedure 52(a)(6) commands that a reviewing court “must not … set aside” a district court’s “[f]indings of fact” unless they are “clearly erroneous.” That rule, the Court emphasized, contains no exception for patent cases. Nothing in Markman purported to create one; Markman allocated construction to the judge rather than the jury, but it did not convert every factual premise of a construction into a question of law immune from deference.

Justice Breyer’s opinion drew a careful line. The ultimate question—the legal meaning the court assigns to a claim term—remains a conclusion of law reviewed de novo. So too, when a district court construes a claim using only the intrinsic record (the claims, specification, and prosecution history), the exercise is “solely a determination of law” reviewed de novo, because reading those documents is the kind of textual interpretation appellate courts perform without deference. But when the court must look beyond the patent to extrinsic evidence and resolve a subsidiary dispute of fact—say, by crediting one expert’s account of how artisans use a term—that finding is reviewed for clear error. The appellate court then reviews de novo whether the trial judge’s construction, given those facts, was legally correct.

Why deference fits the trial court’s role

The Court’s rationale was institutional. A trial judge who presides over a Markman hearing, hears experts testify, and assesses their credibility develops an expertise about the specific technical dispute that an appellate panel reading a cold record cannot replicate. That comparative advantage is precisely why Rule 52 demands deference to fact-finding generally, and the Court saw no reason patent cases should be different. Clear-error review, Breyer reasoned, would also reduce the notorious instability of claim construction on appeal, where high reversal rates under Cybor had made district-court rulings feel provisional.

The dissent, by Justice Thomas, resisted the line-drawing. In its view, claim construction is more akin to interpreting a statute or other legal instrument—an exercise that may use evidence about specialized meaning without thereby becoming “fact-finding” subject to Rule 52. The disagreement exposes a genuine ambiguity at the heart of claim construction: it is neither purely legal interpretation nor purely factual determination, but an uneasy hybrid the Court tried to disaggregate.

Open questions

Teva drew a boundary that proves slippery in practice. Many constructions blend intrinsic and extrinsic analysis, and characterizing a given step as “factual” (clear error) or “legal” (de novo) can itself be outcome-determinative—inviting strategic framing by both sides. The decision also did little to change results in the case that produced it: on remand, the Federal Circuit again held the relevant claims indefinite, illustrating that deference to fact-finding does not always rescue a construction. And because Phillips steers courts toward the intrinsic record—which Teva treats as a matter of law—litigants and judges now have an incentive to keep disputes “intrinsic,” leaving unsettled how much extrinsic fact-finding actually triggers deferential review.

Implications

  • Build the factual record at trial. Subsidiary findings supported by credited expert testimony are now shielded by clear-error review, so the Markman hearing is where battles over technical meaning should be won.
  • Frame the dispute deliberately. Whether a construction rests on intrinsic documents (de novo) or extrinsic facts (clear error) can decide the appeal; both sides will litigate that characterization.
  • Intrinsic-only constructions still get fresh review. Where the patent’s own documents resolve a term, the Federal Circuit reviews de novo, as before.
  • Appeals got harder for losers below. Deference to fact-finding raises the bar for overturning a construction grounded in live testimony.
  • The hybrid nature of construction endures. Teva disaggregated law from fact but did not dissolve the tension the dissent identified.

Frequently asked questions

Did Teva overrule Markman? No. Markman still governs who construes claims—the judge. Teva addressed a different question: how an appellate court reviews the judge’s work, holding that subsidiary factual findings get clear-error deference while the ultimate construction is reviewed de novo.

When does clear-error review apply after Teva? Only to subsidiary factual findings the district court makes when it consults extrinsic evidence—such as crediting an expert about how skilled artisans understand a term. Constructions based solely on the patent’s intrinsic record are reviewed de novo.

Did Teva ultimately win its case? No. The Supreme Court vacated and remanded for review under the correct standard, but on remand the Federal Circuit again held the disputed claims indefinite—showing that deferential review does not guarantee a different outcome.

Authorities and sources