Patents

The Specification Comes First: Phillips v. AWH Corp. and the Hierarchy of Claim-Construction Evidence

Sitting en banc, the Federal Circuit demoted the dictionary and elevated the patent's own specification as the single best guide to claim meaning.

May 19, 2025
Open technical patent document with engineering drawings and a magnifying glass
Phillips made the patent's specification, not a general-purpose dictionary, the primary lens for reading claim terms. 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 Phillips v. AWH Corp., 415 F.3d 1303 (Fed. Cir. 2005) (en banc) (Nos. 03-1269, -1286), the U.S. Court of Appeals for the Federal Circuit took up claim construction in its most ambitious form and tried to impose order on a field that had grown chaotic. Decided July 12, 2005, in an opinion by Judge Bryson, the en banc court answered the methodological question Markman had left open: once the judge is the one reading the claims, what does the judge read, and in what order? The answer—intrinsic evidence first, with the specification as the single best guide—remains the governing framework two decades later and is cited in nearly every modern claim-construction dispute.

At a glance

  • Court and date: U.S. Court of Appeals for the Federal Circuit, en banc, decided July 12, 2005.
  • Docket: Nos. 03-1269, -1286.
  • Patent: U.S. Patent No. 4,677,798, directed to modular steel-shell wall panels that could be assembled into vandalism- and impact-resistant structures.
  • Disputed term: “baffles”—internal steel members “extending inwardly from the steel shell walls.” The fight was whether the term was limited to baffles oriented at angles other than 90 degrees, as the preferred embodiments depicted.
  • Core question: What is the proper relationship among the claims, the specification, the prosecution history, and outside sources such as dictionaries and expert testimony?
  • Holding: Intrinsic evidence controls; the specification is “the single best guide” to a disputed term’s meaning; dictionaries may help but cannot trump the patent’s own usage.

The problem Phillips set out to fix

By 2005 the Federal Circuit’s claim-construction case law had split into two camps. One line, exemplified by Texas Digital Systems, Inc. v. Telegenix, Inc., 308 F.3d 1193 (Fed. Cir. 2002), began with dictionaries: a court would first collect the ordinary dictionary meanings of a term and treat them as presumptively correct, consulting the specification mainly to see whether the patentee had disclaimed or redefined the word. A competing line treated the specification as primary from the outset. The two methodologies could yield opposite results from the same patent, and litigants gamed the difference.

Phillips was the en banc court’s chance to choose. The “baffles” dispute was a clean vehicle: a general-purpose dictionary defined a baffle broadly as something that deflects or regulates flow, while the ‘798 patent repeatedly described baffles deployed at oblique angles to deflect projectiles. A dictionary-first approach pointed toward a broad construction; a specification-first approach pointed toward something narrower. The court used the case to declare which discipline governs.

Intrinsic evidence first

The court’s central holding is a hierarchy. Claim construction begins with the words of the claims themselves, read in the context of the entire patent and given the ordinary meaning they would carry for a person of ordinary skill in the art at the time of the invention. That person, crucially, does not read a claim term in isolation; the artisan reads it against the backdrop of the whole patent.

From there the court ranked the sources. The specification is “always highly relevant” and is, in the court’s now-famous phrase, “the single best guide to the meaning of a disputed term.” The prosecution history—the record of the patentee’s dealings with the Patent Office—comes next, useful for showing how the applicant understood the invention and what it may have disclaimed. These intrinsic sources, generated by the patentee and the examiner and available to the public, are the materials a competitor would consult to determine a patent’s scope.

Extrinsic evidence—dictionaries, treatises, and expert testimony—occupies the bottom tier. The court did not banish it. Dictionaries can help a judge understand the technology, and expert testimony can explain how artisans use terms. But extrinsic sources are “less reliable” than the intrinsic record: they are not part of the patent, are not created with the discipline of prosecution, and can be marshaled by litigants to manufacture ambiguity. They may not be used to contradict a meaning that is clear from the intrinsic evidence.

Rejecting the dictionary-first method

The most consequential move in Phillips was its explicit repudiation of Texas Digital’s methodology. Beginning with a dictionary, the court warned, “focuses the inquiry on the abstract meaning of words rather than on the meaning of claim terms within the context of the patent.” A general-purpose dictionary catalogs every plausible sense of a word; a patent uses a word in one particular sense, defined by the invention the applicant actually described. To privilege the dictionary is to risk importing meanings the patentee never intended and a competitor would never have inferred.

At the same time, the court was careful to guard against the opposite error: reading limitations from the specification’s preferred embodiments into the claims. The line Phillips drew—between using the specification to interpret a claim term and improperly using it to import a limitation—is notoriously fine, and the opinion candidly acknowledged that “there is no magic formula” for walking it. The court also disclaimed any rigid sequence, holding that judges need not consult the sources in a fixed order so long as they do not let extrinsic evidence override an unambiguous intrinsic meaning. Applying its framework, the en banc court read “baffles” more broadly than the panel had and remanded.

Open questions

Phillips gave courts a hierarchy but not an algorithm. Its own admission that there is “no magic formula” for distinguishing legitimate interpretation from improper importation of limitations means the hardest cases still turn on judgment calls that reasonable jurists decide differently. The opinion also assumed a coherent intrinsic record; it offers less guidance when the specification is silent, internally inconsistent, or written in terms a lay judge cannot parse without the very expert testimony Phillips demotes. And by framing construction as a largely intrinsic, legal exercise, Phillips reinforced the de novo review posture that the Supreme Court would later qualify in Teva v. Sandoz—leaving an unresolved tension between Phillips’s methodology and Teva’s deference to fact-finding.

Implications

  • Win on the intrinsic record. The specification and prosecution history are the highest-value materials; arguments grounded there beat arguments grounded in dictionaries or experts.
  • Draft the specification as a glossary. Because the specification is the single best guide, patentees should describe terms with the breadth—or precision—they want courts to enforce.
  • Dictionaries are support, not foundation. Extrinsic evidence can illuminate but cannot contradict a meaning clear from the patent itself.
  • Mind the embodiment trap. Litigants must distinguish using the specification to interpret a term from improperly confining the claim to its preferred embodiments.
  • No mandatory sequence. Courts may weigh the sources flexibly, provided extrinsic evidence never overrides unambiguous intrinsic meaning.

Frequently asked questions

What is the single most important takeaway from Phillips? That the patent’s own specification—not a general-purpose dictionary—is the primary guide to what a disputed claim term means, with the claims, specification, and prosecution history (the intrinsic record) controlling over outside sources.

Did Phillips ban the use of dictionaries and expert testimony? No. It demoted them. Extrinsic evidence remains admissible to help a court understand the technology, but it cannot be used to contradict a meaning that is clear from the intrinsic record.

How does Phillips avoid simply reading the claims as narrowly as the examples? It draws a line between using the specification to interpret a term and improperly importing limitations from preferred embodiments. The court conceded the line is difficult and that no formula resolves every case.

Authorities and sources