WILLIAMS v. TAYLOR

TERRY WILLIAMS, PETITIONER v. JOHN TAYLOR, WARDEN

on writ of certiorari to the united states court of appeals for the fourth circuit

---

---

 Justice Stevens announced the judgment of the Court and delivered the opinion of the Court with respect to Parts I, III, and IV, and an opinion with respect to Parts II and V.^**

 The questions presented are whether Terry Williams’ constitutional right to the effective assistance of counsel as defined in Strickland v. Washington, 466 U. S. 668 (1984), was violated, and whether the judgment of the Virginia Supreme Court refusing to set aside his death sentence “was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States,” within the meaning of 28 U. S. C. §2254(d)(1) (1994 ed., Supp. III). We answer both questions affirmatively.

I

 On November 3, 1985, Harris Stone was found dead in his residence on Henry Street in Danville, Virginia. Finding no indication of a struggle, local officials determined that the cause of death was blood alcohol poisoning, and the case was considered closed. Six months after Stone’s death, Terry Williams, who was then incarcerated in the “I” unit of the city jail for an unrelated offense, wrote a letter to the police stating that he had killed “   ‘that man down on Henry Street’   ” and also stating that he “   ‘did it’   ” to that “   ‘lady down on West Green Street’   ” and was “   ‘very sorry.’   ” The letter was unsigned, but it closed with a reference to “I cell.” App. 41. The police readily identified Williams as its author, and, on April 25, 1986, they obtained several statements from him. In one Williams admitted that, after Stone refused to lend him “   ‘a couple of dollars,’   ” he had killed Stone with a mattock and took the money from his wallet.²Id., at 4. In September 1986, Williams was convicted of robbery and capital murder.

 At Williams’ sentencing hearing, the prosecution proved that Williams had been convicted of armed robbery in 1976 and burglary and grand larceny in 1982. The prosecution also introduced the written confessions that Williams had made in April. The prosecution described two auto thefts and two separate violent assaults on elderly victims perpetrated after the Stone murder. On December 4, 1985, Williams had started a fire outside one victim’s residence before attacking and robbing him. On March 5, 1986, Williams had brutally assaulted an elderly woman on West Green Street—an incident he had mentioned in his letter to the police. That confession was particularly damaging because other evidence established that the woman was in a “vegetative state” and not expected to recover. Id., at 60. Williams had also been convicted of arson for setting a fire in the jail while awaiting trial in this case. Two expert witnesses employed by the State testified that there was a “high probability” that Williams would pose a serious continuing threat to society. Id., at 89.

 The evidence offered by Williams’ trial counsel at the sentencing hearing consisted of the testimony of Williams’ mother, two neighbors, and a taped excerpt from a statement by a psychiatrist. One of the neighbors had not been previously interviewed by defense counsel, but was noticed by counsel in the audience during the proceedings and asked to testify on the spot. The three witnesses briefly described Williams as a “nice boy” and not a violent person. Id., at 124. The recorded psychiatrist’s testimony did little more than relate Williams’ statement during an examination that in the course of one of his earlier robberies, he had removed the bullets from a gun so as not to injure anyone.

 In his cross-examination of the prosecution witnesses, Williams’ counsel repeatedly emphasized the fact that Williams had initiated the contact with the police that enabled them to solve the murder and to identify him as the perpetrator of the recent assaults, as well as the car thefts. In closing argument, Williams’ counsel characterized Williams’ confessional statements as “dumb,” but asked the jury to give weight to the fact that he had “turned himself in, not on one crime but on four … that the [police otherwise] would not have solved.” Id., at 140. The weight of defense counsel’s closing, however, was devoted to explaining that it was difficult to find a reason why the jury should spare Williams’ life.³

 The jury found a probability of future dangerousness and unanimously fixed Williams’ punishment at death. The trial judge concluded that such punishment was “proper” and “just” and imposed the death sentence. Id., at 154. The Virginia Supreme Court affirmed the conviction and sentence. Williams v. Commonwealth, 234 Va. 168, 360 S. E. 2d 361 (1987), cert. denied, Williams v. Virginia, 484 U. S. 1020 (1988). It rejected Williams’ argument that when the trial judge imposed sentence, he failed to give mitigating weight to the fact that Williams had turned himself in. 234 Va., at 181–182, 360 S. E. 2d, at 369–370.

State Habeas Corpus Proceedings

 In 1988 Williams filed for state collateral relief in the Danville Circuit Court. The petition was subsequently amended, and the Circuit Court (the same judge who had presided over Williams’ trial and sentencing) held an evidentiary hearing on Williams’ claim that trial counsel had been ineffective.⁴ Based on the evidence adduced after two days of hearings, Judge Ingram found that Williams’ conviction was valid, but that his trial attorneys had been ineffective during sentencing. Among the evidence reviewed that had not been presented at trial were documents prepared in connection with Williams’ commitment when he was 11 years old that dramatically described mistreatment, abuse, and neglect during his early childhood, as well as testimony that he was “borderline mentally retarded,” had suffered repeated head injuries, and might have mental impairments organic in origin. App. 528–529, 595. The habeas hearing also revealed that the same experts who had testified on the State’s behalf at trial believed that Williams, if kept in a “structured environment,” would not pose a future danger to society. Id., at 313–314.

 Counsel’s failure to discover and present this and other significant mitigating evidence was “below the range expected of reasonable, professional competent assistance of counsel.” Id., at 424. Counsels’ performance thus “did not measure up to the standard required under the holding of Strickland v. Washington, 466 U. S. 668 (1984), and [if it had,] there is a reasonable probability that the result of the sentencing phase would have been different.” Id., at 429. Judge Ingram therefore recommended that Williams be granted a rehearing on the sentencing phase of his trial.

 The Virginia Supreme Court did not accept that recommendation. Williams v. Warden, 254 Va. 16, 487 S. E. 2d 194 (1997). Although it assumed, without deciding, that trial counsel had been ineffective, id., at 23–26, 487 S. E. 2d, at 198, 200, it disagreed with the trial judge’s conclusion that Williams had suffered sufficient prejudice to warrant relief. Treating the prejudice inquiry as a mixed question of law and fact, the Virginia Supreme Court accepted the factual determination that available evidence in mitigation had not been presented at the trial, but held that the trial judge had misapplied the law in two respects. First, relying on our decision in Lockhart v. Fretwell, 506 U. S. 364 (1993), the court held that it was wrong for the trial judge to rely “   ‘on mere outcome determination’   ” when assessing prejudice, 254 Va., at 23, 487 S. E. 2d, at 198 (quoting Lockhart, 506 U. S., at 369). Second, it construed the trial judge’s opinion as having “adopted a per se approach” that would establish prejudice whenever any mitigating evidence was omitted. 254 Va., at 26, 487 S. E. 2d, at 200.

 The court then reviewed the prosecution evidence supporting the “future dangerousness” aggravating circumstance, reciting Williams’ criminal history, including the several most recent offenses to which he had confessed. In comparison, it found that the excluded mitigating evidence—which it characterized as merely indicating “that numerous people, mostly relatives, thought that defendant was nonviolent and could cope very well in a structured environment,” ibid.—“barely would have altered the profile of this defendant that was presented to the jury,” ibid. On this basis, the court concluded that there was no reasonable possibility that the omitted evidence would have affected the jury’s sentencing recommendation, and that Williams had failed to demonstrate that his sentencing proceeding was fundamentally unfair.

Federal Habeas Corpus Proceedings

 Having exhausted his state remedies, Williams sought a federal writ of habeas corpus pursuant to 28 U. S. C. §2254 (1994 ed. and Supp. III). After reviewing the state habeas hearing transcript and the state courts’ findings of fact and conclusions of law, the federal trial judge agreed with the Virginia trial judge: The death sentence was constitutionally infirm.

 After noting that the Virginia Supreme Court had not addressed the question whether trial counsel’s performance at the sentencing hearing fell below the range of competence demanded of lawyers in criminal cases, the judge began by addressing that issue in detail. He identified five categories of mitigating evidence that counsel had failed to introduce,⁵ and he rejected the argument that counsel’s failure to conduct an adequate investigation had been a strategic decision to rely almost entirely on the fact that Williams had voluntarily confessed.

 According to Williams’ trial counsel’s testimony before the state habeas court, counsel did not fail to seek Williams’ juvenile and social services records because he thought they would be counterproductive, but because counsel erroneously believed that “   ‘state law didn’t permit it.’   ” App. 470. Counsel also acknowledged in the course of the hearings that information about Williams’ childhood would have been important in mitigation. And counsel’s failure to contact a potentially persuasive character witness was likewise not a conscious strategic choice, but simply a failure to return that witness’ phone call offering his service. Id., at 470–471. Finally, even if counsel neglected to conduct such an investigation at the time as part of a tactical decision, the District Judge found, tactics as a matter of reasonable performance could not justify the omissions.

 Turning to the prejudice issue, the judge determined that there was “   ‘a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.’ Strickland, 466 U. S. at 694.” Id., at 473. He found that the Virginia Supreme Court had erroneously assumed that Lockhart had modified the Strickland standard for determining prejudice, and that it had made an important error of fact in discussing its finding of no prejudice.⁶ Having introduced his analysis of Williams’ claim with the standard of review applicable on habeas appeals provided by 28 U. S. C. §2254(d) (1994 ed., Supp. III), the judge concluded that those errors established that the Virginia Supreme Court’s decision “was contrary to, or involved an unreasonable application of, clearly established Federal law” within the meaning of §2254(d)(1).

 The Federal Court of Appeals reversed. 163 F. 3d 860 (CA4 1998). It construed §2254(d)(1) as prohibiting the grant of habeas corpus relief unless the state court “   ‘decided the question by interpreting or applying the relevant precedent in a manner that reasonable jurists would all agree is unreasonable.’   ” Id., at 865 (quoting Green v. French, 143 F. 3d 865, 870 (CA4 1998)). Applying that standard, it could not say that the Virginia Supreme Court’s decision on the prejudice issue was an unreasonable application of the tests developed in either Strickland or Lockhart.7 It explained that the evidence that Williams presented a future danger to society was “simply overwhelming,” 163 F. 3d, at 868, it endorsed the Virginia Supreme Court’s interpretation of Lockhart, 163 F. 3d, at 869, and it characterized the state court’s understanding of the facts in this case as “reasonable,” id., at 870.

 We granted certiorari, 526 U. S. 1050 (1999), and now reverse.

II

 In 1867, Congress enacted a statute providing that federal courts “shall have power to grant writs of habeas corpus in all cases where any person may be restrained of his or her liberty in violation of the constitution, or of any treaty or law of the United States . …” Act of Feb. 5, 1867, ch. 28, §1, 14 Stat. 385. Over the years, the federal habeas corpus statute has been repeatedly amended, but the scope of that jurisdictional grant remains the same.⁸ It is, of course, well settled that the fact that constitutional error occurred in the proceedings that led to a state-court conviction may not alone be sufficient reason for concluding that a prisoner is entitled to the remedy of habeas. See, e.g., Stone v. Powell, 428 U. S. 465 (1976); Brecht v. Abrahamson, 507 U. S. 619 (1993). On the other hand, errors that undermine confidence in the fundamental fairness of the state adjudication certainly justify the issuance of the federal writ. See, e.g., Teague v. Lane, 489 U. S. 288, 311–314 (1989) (quoting Mackey v. United States, 401 U. S. 667, 692–694 (1971) (opinion of Harlan, J., concurring in judgments in part and dissenting in part), and quoting Rose v. Lundy, 455 U. S 509, 544 (1982) (Stevens, J., dissenting)). The deprivation of the right to the effective assistance of counsel recognized in Strickland is such an error. Strickland, 466 U. S., at 686, 697–698.

 The warden here contends that federal habeas corpus relief is prohibited by the amendment to 28 U. S. C. §2254 (1994 ed., Supp. III), enacted as a part of the Antiterrorism and Effective Death Penalty Act of 1996 (AEDPA). The relevant portion of that amendment provides:

 “(d)  An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim—

 “(1)  resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States; … .”

 In this case, the Court of Appeals applied the construction of the amendment that it had adopted in its earlier opinion in Green v. French, 143 F. 3d 865 (CA4 1998). It read the amendment as prohibiting federal courts from issuing the writ unless:

 “   ‘(a)  the state court decision is in ‘square conflict’ with Supreme Court precedent that is controlling as to law and fact or (b) if no such controlling decision exists, ‘the state court’s resolution of a question of pure law rests upon an objectively unreasonable derivation of legal principles from the relevant [S]upreme [C]ourt precedents, or if its decision rests upon an objectively unreasonable application of established principles to new facts,’   ” 163 F. 3d, at 865 (quoting Green, 143 F. 3d, at 870).

Accordingly, it held that a federal court may issue habeas relief only if “the state courts have decided the question by interpreting or applying the relevant precedent in a manner that reasonable jurists would all agree is unreasonable,” 163 F. 3d, at 865.9

 We are convinced that that interpretation of the amendment is incorrect. It would impose a test for determining when a legal rule is clearly established that simply cannot be squared with the real practice of decisional law.¹⁰ It would apply a standard for determining the “reasonableness” of state-court decisions that is not contained in the statute itself, and that Congress surely did not intend. And it would wrongly require the federal courts, including this Court, to defer to state judges’ interpretations of federal law.

 As the Fourth Circuit would have it, a state-court judgment is “unreasonable” in the face of federal law only if all reasonable jurists would agree that the state court was unreasonable. Thus, in this case, for example, even if the Virginia Supreme Court misread our opinion in Lockhart, we could not grant relief unless we believed that none of the judges who agreed with the state court’s interpretation of that case was a “reasonable jurist.” But the statute says nothing about “reasonable judges,” presumably because all, or virtually all, such judges occasionally commit error; they make decisions that in retrospect may be characterized as “unreasonable.” Indeed, it is most unlikely that Congress would deliberately impose such a requirement of unanimity on federal judges. As Congress is acutely aware, reasonable lawyers and lawgivers regularly disagree with one another. Congress surely did not intend that the views of one such judge who might think that relief is not warranted in a particular case should always have greater weight than the contrary, considered judgment of several other reasonable judges.

 The inquiry mandated by the amendment relates to the way in which a federal habeas court exercises its duty to decide constitutional questions; the amendment does not alter the underlying grant of jurisdiction in §2254(a), see n. 7, supra.¹¹ When federal judges exercise their federal-question jurisdiction under the “judicial Power” of Article III of the Constitution, it is “emphatically the province and duty” of those judges to “say what the law is.” Marbury v. Madison, 1 Cranch 137, 177 (1803). At the core of this power is the federal courts’ independent responsibility—independent from its coequal branches in the Federal Government, and independent from the separate authority of the several States—to interpret federal law. A construction of AEDPA that would require the federal courts to cede this authority to the courts of the States would be inconsistent with the practice that federal judges have traditionally followed in discharging their duties under Article III of the Constitution. If Congress had intended to require such an important change in the exercise of our jurisdiction, we believe it would have spoken with much greater clarity than is found in the text of AEDPA.

 This basic premise informs our interpretation of both parts of §2254(d)(1): first, the requirement that the determinations of state courts be tested only against “clearly established Federal law, as determined by the Supreme Court of the United States,” and second, the prohibition on the issuance of the writ unless the state court’s decision is “contrary to, or involved an unreasonable application of,” that clearly established law. We address each part in turn.

The “clearly established law” requirement

 In Teague v. Lane, 489 U. S. 288 (1989), we held that the petitioner was not entitled to federal habeas relief because he was relying on a rule of federal law that had not been announced until after his state conviction became final. The antiretroactivity rule recognized in Teague, which prohibits reliance on “new rules,” is the functional equivalent of a statutory provision commanding exclusive reliance on “clearly established law.” Because there is no reason to believe that Congress intended to require federal courts to ask both whether a rule sought on habeas is “new” under Teague—which remains the law—and also whether it is “clearly established” under AEDPA, it seems safe to assume that Congress had congruent concepts in mind.¹² It is perfectly clear that AEDPA codifies Teague to the extent that Teague requires federal habeas courts to deny relief that is contingent upon a rule of law not clearly established at the time the state conviction became final.¹³

 Teague’s core principles are therefore relevant to our construction of this requirement. Justice Harlan recognized the “inevitable difficulties” that come with “attempting ‘to determine whether a particular decision has really announced a “new” rule at all or whether it has simply applied a well-established constitutional principle to govern a case which is closely analogous to those which have been previously considered in the prior case law.’   ” Mackey, 401 U. S., at 695 (quoting Desist v. United States, 394 U. S. 244, 263 (1969)). But Teague established some guidance for making this determination, explaining that a federal habeas court operates within the bounds of comity and finality if it applies a rule “dictated by precedent existing at the time the defendant’s conviction became final.” Teague, 489 U. S., at 301 (emphasis deleted). A rule that “breaks new ground or imposes a new obligation on the States or the Federal Government,” ibid., falls outside this universe of federal law.

 To this, AEDPA has added, immediately following the “clearly established law” requirement, a clause limiting the area of relevant law to that “determined by the Supreme Court of the United States.” 28 U. S. C. §2254(d)(1) (1994 ed., Supp. III). If this Court has not broken sufficient legal ground to establish an asked-for constitutional principle, the lower federal courts cannot themselves establish such a principle with clarity sufficient to satisfy the AEDPA bar. In this respect, we agree with the Seventh Circuit that this clause “extends the principle of Teague by limiting the source of doctrine on which a federal court may rely in addressing the application for a writ.” Lindh v. Murphy, 96 F. 3d 856, 869 (CA7 1996). As that court explained:

“This is a retrenchment from former practice, which allowed the United States courts of appeals to rely on their own jurisprudence in addition to that of the Supreme Court. The novelty in this portion §2254(d)(1) is not the ‘contrary to’ part but the reference to ‘Federal law, as determined by the Supreme Court of the United States’ (emphasis added). This extends the principle of Teague [v. Lane, 489 U. S. 288 (1989)] by limiting the source of doctrine on which a federal court may rely in addressing the application for a writ. It does not, however, purport to limit the federal courts’ independent interpretive authority with respect to federal questions.” Ibid.

A rule that fails to satisfy the foregoing criteria is barred by Teague from application on collateral review, and, similarly, is not available as a basis for relief in a habeas case to which AEDPA applies.

 In the context of this case, we also note that, as our precedent interpreting Teague has demonstrated, rules of law may be sufficiently clear for habeas purposes even when they are expressed in terms of a generalized standard rather than as a bright-line rule. As Justice Kennedy has explained:

 “If the rule in question is one which of necessity requires a case-by-case examination of the evidence, then we can tolerate a number of specific applications without saying that those applications themselves create a new rule… . Where the beginning point is a rule of this general application, a rule designed for the specific purpose of evaluating a myriad of factual contexts, it will be the infrequent case that yields a result so novel that it forges a new rule, one not dictated by precedent.” Wright v. West, 505 U. S. 277, 308–309 (1992) (opinion concurring in judgment).

Moreover, the determination whether or not a rule is clearly established at the time a state court renders its final judgment of conviction is a question as to which the “federal courts must make an independent evaluation.” Id., at 305 (O’Connor, J., concurring in judgment); accord id., at 307 (Kennedy, J., concurring).

 It has been urged, in contrast, that we should read Teague and its progeny to encompass a broader principle of deference requiring federal courts to “validat[e] ‘reasonable, good-faith interpretations’ of the law” by state courts. Brief for California et al. as Amici Curiae 6 (quoting Butler v. McKellar, 494 U. S. 407, 414 (1990)). The position has been bolstered with references to our statements elucidating the ‘new rule’ inquiry as one turning on whether “reasonable jurists” would agree the rule was not clearly established. Sawyer v. Smith, 497 U. S. 227, 234 (1990). This presumption of deference was in essence the position taken by three Members of this Court in Wright, 505 U. S., at 290–291 (opinion of Thomas, J.) (“[A] federal habeas court ‘must defer to the state court’s decision rejecting the claim unless that decision is patently unreasonable’   ”) (quoting Butler, 494 U. S., at 422 (Brennan, J., dissenting)).

 Teague, however, does not extend this far. The often repeated language that Teague endorses “reasonable, good-faith interpretations” by state courts is an explanation of policy, not a statement of law. The Teague cases reflect this Court’s view that habeas corpus is not to be used as a second criminal trial, and federal courts are not to run roughshod over the considered findings and judgments of the state courts that conducted the original trial and heard the initial appeals. On the contrary, we have long insisted that federal habeas courts attend closely to those considered decisions, and give them full effect when their findings and judgments are consistent with federal law. See Thompson v. Keohane, 516 U. S. 99, 107–116 (1995). But as Justice O’Connor explained in Wright:

 “[T]he duty of the federal court in evaluating whether a rule is ‘new’ is not the same as deference; … Teague does not direct federal courts to spend less time or effort scrutinizing the existing federal law, on the ground that they can assume the state courts interpreted it properly.

 “[T]he maxim that federal courts should ‘give great weight to the considered conclusions of a coequal state judiciary’ … does not mean that we have held in the past that federal courts must presume the correctness of a state court’s legal conclusions on habeas, or that a state court’s incorrect legal determination has ever been allowed to stand because it was reasonable. We have always held that federal courts, even on habeas, have an independent obligation to say what the law is.” 505 U. S., at 305.

 We are convinced that in the phrase, “clearly established law,” Congress did not intend to modify that independent obligation.

The “contrary to, or an unreasonable application of,” requirement

 The message that Congress intended to convey by using the phrases, “contrary to” and “unreasonable application of” is not entirely clear. The prevailing view in the Circuits is that the former phrase requires de novo review of ‘pure’ questions of law and the latter requires some sort of “reasonability” review of so-called mixed questions of law and fact. See, e.g., Neelley v. Nagle, 138 F. 3d 917 (CA11 1998); Drinkard v. Johnson, 97 F. 3d 751 (CA5 1996); Lindh v. Murphy, 96 F. 3d 856 (CA7 1996) (en banc), rev’d on other grounds, 521 U. S. 320 (1997).

 We are not persuaded that the phrases define two mutually exclusive categories of questions. Most constitutional questions that arise in habeas corpus proceedings—and therefore most “decisions” to be made—require the federal judge to apply a rule of law to a set of facts, some of which may be disputed and some undisputed. For example, an erroneous conclusion that particular circumstances established the voluntariness of a confession, or that there exists a conflict of interest when one attorney represents multiple defendants, may well be described either as “contrary to” or as an “unreasonable application of   ” the governing rule of law. Cf. Miller v. Fenton, 474 U. S. 104, 116 (1985); Cuyler v. Sullivan, 446 U. S. 335, 341–342 (1980). In constitutional adjudication, as in the common law, rules of law often develop incrementally as earlier decisions are applied to new factual situations. See Wright, 505 U. S., at 307 (Kennedy, J., concurring). But rules that depend upon such elaboration are hardly less lawlike than those that establish a bright-line test.

 Indeed, our pre-AEDPA efforts to distinguish questions of fact, questions of law, and “mixed questions,” and to create an appropriate standard of habeas review for each, generated some not insubstantial differences of opinion as to which issues of law fell into which category of question, and as to which standard of review applied to each. See Thompson, 516 U. S., at 110–111 (acknowledging “   ‘that the Court has not charted an entirely clear course in this area’   ” and that “the proper characterization of a question as one of fact or law is sometimes slippery”) (quoting Miller, 474 U. S., at 113). We thus think the Fourth Circuit was correct when it attributed the lack of clarity in the statute, in part, to the overlapping meanings of the phrases “contrary to” and “unreasonable application of.” See Green, 143 F. 3d, at 870.

 The statutory text likewise does not obviously prescribe a specific, recognizable standard of review for dealing with either phrase. Significantly, it does not use any term, such as “de novo” or “plain error,” that would easily identify a familiar standard of review. Rather, the text is fairly read simply as a command that a federal court not issue the habeas writ unless the state court was wrong as a matter of law or unreasonable in its application of law in a given case. The suggestion that a wrong state-court “decision”—a legal judgment rendered “after consideration of facts, and … law,” Black’s Law Dictionary 407 (6th ed. 1990) (emphasis added)—may no longer be redressed through habeas (because it is unreachable under the “unreasonable application” phrase) is based on a mistaken insistence that the §2254(d)(1) phrases have not only independent, but mutually exclusive, meanings. Whether or not a federal court can issue the writ “under [the] ‘unreasonable application’ clause,” the statute is clear that habeas may issue under §2254(d)(1) if a state court “decision” is “contrary to … clearly established Federal law.” We thus anticipate that there will be a variety of cases, like this one, in which both phrases may be implicated.

 Even though we cannot conclude that the phrases establish “a body of rigid rules,” they do express a “mood” that the federal judiciary must respect. Universal Camera Corp. v. NLRB, 340 U. S. 474, 487 (1951). In this respect, it seems clear that Congress intended federal judges to attend with the utmost care to state-court decisions, including all of the reasons supporting their decisions, before concluding that those proceedings were infected by constitutional error sufficiently serious to warrant the issuance of the writ. Likewise, the statute in a separate provision provides for the habeas remedy when a state-court decision “was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.” 28 U. S. C. §2254(d)(2) (1994 ed., Supp. III) (emphasis added). While this provision is not before us in this case, it provides relevant context for our interpretation of §2254(d)(1); in this respect, it bolsters our conviction that federal habeas courts must make as the starting point of their analysis the state courts’ determinations of fact, including that aspect of a “mixed question” that rests on a finding of fact. AEDPA plainly sought to ensure a level of “deference to the determinations of state courts,” provided those determinations did not conflict with federal law or apply federal law in an unreasonable way. H. R. Conf. Rep. No. 104–518, p. 111 (1996). Congress wished to curb delays, to prevent “retrials” on federal habeas, and to give effect to state convictions to the extent possible under law. When federal courts are able to fulfill these goals within the bounds of the law, AEDPA instructs them to do so.

 On the other hand, it is significant that the word “deference” does not appear in the text of the statute itself. Neither the legislative history, nor the statutory text, suggests any difference in the so-called “deference” depending on which of the two phrases is implicated.¹⁴ Whatever “deference” Congress had in mind with respect to both phrases, it surely is not a requirement that federal courts actually defer to a state-court application of the federal law that is, in the independent judgment of the federal court, in error. As Judge Easterbrook noted with respect to the phrase “contrary to”:

“Section 2254(d) requires us to give state courts’ opinions a respectful reading, and to listen carefully to their conclusions, but when the state court addresses a legal question, it is the law ‘as determined by the Supreme Court of the United States’ that prevails.” Lindh, 96 F. 3d, at 869.¹⁵

 Our disagreement with Justice O’Connor about the precise meaning of the phrase “contrary to,” and the word “unreasonable,” is, of course, important, but should affect only a narrow category of cases. The simplest and first definition of “contrary to” as a phrase is “in conflict with.” Webster’s Ninth New Collegiate Dictionary 285 (1983). In this sense, we think the phrase surely capacious enough to include a finding that the state-court “decision” is simply “erroneous” or wrong. (We hasten to add that even “diametrically different” from, or “opposite” to, an established federal law would seem to include “decisions” that are wrong in light of that law.) And there is nothing in the phrase “contrary to”—as Justice O’Connor appears to agree—that implies anything less than independent review by the federal courts.  Moreover, state-court decisions that do not “conflict” with federal law will rarely be “unreasonable” under either her reading of the statute or ours. We all agree that state-court judgments must be upheld unless, after the closest examination of the state-court judgment, a federal court is firmly convinced that a  

federal constitutional right has been violated. Our difference is as to the cases in which, at first-blush, a state-court judgment seems entirely reasonable, but thorough analysis by a federal court produces a firm conviction that that judgment is infected by constitutional error. In our view, such an erroneous judgment is “unreasonable” within the meaning of the act even though that conclusion was not immediately apparent.

 In sum, the statute directs federal courts to attend to every state-court judgment with utmost care, but it does not require them to defer to the opinion of every reasonable state-court judge on the content of federal law. If, after carefully weighing all the reasons for accepting a state court’s judgment, a federal court is convinced that a prisoner’s custody—or, as in this case, his sentence of death—violates the Constitution, that independent judgment should prevail. Otherwise the federal “law as determined by the Supreme Court of the United States” might be applied by the federal courts one way in Virginia and another way in California. In light of the well-recognized interest in ensuring that federal courts interpret federal law in a uniform way,¹⁶ we are convinced that Congress did not intend the statute to produce such a result.

III

 In this case, Williams contends that he was denied his constitutionally guaranteed right to the effective assistance of counsel when his trial lawyers failed to investigate and to present substantial mitigating evidence to the sentencing jury. The threshold question under AEDPA is whether Williams seeks to apply a rule of law that was clearly established at the time his state-court conviction became final. That question is easily answered because the merits of his claim are squarely governed by our holding in Strickland v. Washington, 466 U. S. 668 (1984).

 We explained in Strickland that a violation of the right on which Williams relies has two components:

“First, the defendant must show that counsel’s performance was deficient. This requires showing that counsel made errors so serious that counsel was not functioning as the ‘counsel’ guaranteed the defendant by the Sixth Amendment. Second, the defendant must show that the deficient performance prejudiced the defense. This requires showing that counsel’s errors were so serious as to deprive the defendant of a fair trial, a trial whose result is reliable.” Id., at 687.

To establish ineffectiveness, a “defendant must show that counsel’s representation fell below an objective standard of reasonableness.” Id., at 688. To establish prejudice he “must show that there is a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different. A reasonable probability is a probability sufficient to undermine confidence in the outcome.” Id., at 694.

 It is past question that the rule set forth in Strickland qualifies as “clearly established Federal law, as determined by the Supreme Court of the United States.” That the Strickland test “of necessity requires a case-by-case examination of the evidence,” Wright, 505 U. S., at 308 (Kennedy, J., concurring), obviates neither the clarity of the rule nor the extent to which the rule must be seen as “established” by this Court. This Court’s precedent “dictated” that the Virginia Supreme Court apply the Strickland test at the time that court entertained Williams’ ineffective-assistance claim. Teague, 489 U. S., at 301. And it can hardly be said that recognizing the right to effective counsel “breaks new ground or imposes a new obligation on the States,” ibid. Williams is therefore entitled to relief if the Virginia Supreme Court’s decision rejecting his ineffective-assistance claim was either “contrary to, or involved an unreasonable application of,” that established law. It was both.

IV

 The Virginia Supreme Court erred in holding that our decision in Lockhart v. Fretwell, 506 U. S. 364 (1993), modified or in some way supplanted the rule set down in Strickland. It is true that while the Strickland test provides sufficient guidance for resolving virtually all ineffective-assistance-of-counsel claims, there are situations in which the overriding focus on fundamental fairness may affect the analysis. Thus, on the one hand, as Strickland itself explained, there are a few situations in which prejudice may be presumed. 466 U. S., at 692. And, on the other hand, there are also situations in which it would be unjust to characterize the likelihood of a different outcome as legitimate “prejudice.” Even if a defendant’s false testimony might have persuaded the jury to acquit him, it is not fundamentally unfair to conclude that he was not prejudiced by counsel’s interference with his intended perjury. Nix v. Whiteside, 475 U. S. 157, 175–176 (1986).

 Similarly, in Lockhart, we concluded that, given the overriding interest in fundamental fairness, the likelihood of a different outcome attributable to an incorrect interpretation of the law should be regarded as a potential “windfall” to the defendant rather than the legitimate “prejudice” contemplated by our opinion in Strickland. The death sentence that Arkansas had imposed on Bobby Ray Fretwell was based on an aggravating circumstance (murder committed for pecuniary gain) that duplicated an element of the underlying felony (murder in the course of a robbery). Shortly before the trial, the United States Court of Appeals for the Eighth Circuit had held that such “double counting” was impermissible, see Collins v. Lockhart, 754 F. 2d 258, 265 (1985), but Fretwell’s lawyer (presumably because he was unaware of the Collins decision) failed to object to the use of the pecuniary gain aggravator. Before Fretwell’s claim for federal habeas corpus relief reached this Court, the Collins case was overruled.¹⁷ Accordingly, even though the Arkansas trial judge probably would have sustained a timely objection to the double counting, it had become clear that the State had a right to rely on the disputed aggravating circumstance. Because the ineffectiveness of Fretwell’s counsel had not deprived him of any substantive or procedural right to which the law entitled him, we held that his claim did not satisfy the “prejudice” component of the Strickland test.¹⁸

 Cases such as Nix v. Whiteside, 475 U. S. 157 (1986), and Lockhart v. Fretwell, 506 U. S. 364 (1993), do not justify a departure from a straightforward application of Strickland when the ineffectiveness of counsel does deprive the defendant of a substantive or procedural right to which the law entitles him.¹⁹ In the instant case, it is undisputed that Williams had a right—indeed, a constitutionally protected right—to provide the jury with the mitigating evidence that his trial counsel either failed to discover or failed to offer.

 Nevertheless, the Virginia Supreme Court read our decision in Lockhart to require a separate inquiry into fundamental fairness even when Williams is able to show that his lawyer was ineffective and that his ineffectiveness probably affected the outcome of the proceeding. It wrote:

 “   ‘The prisoner argues there ‘is a “reasonable probability” that at least one juror would have been moved to spare Petitioner’s life had he heard’ the mitigation evidence developed at the habeas hearing that was not presented at the trial. Summarizing, he contends there ‘is a “reasonable probability” that had at least one juror heard any of this evidence—let alone all of this evidence—the outcome of this case would have been different.’

 “We reject these contentions. The prisoner’s discussion flies in the face of the Supreme Court’s admonition in Lockhart, supra, that ‘an analysis focusing solely on mere outcome determination, without attention to whether the result of the proceeding was fundamentally unfair or unreliable, is defective.’   ” Williams, 254 Va., at 25, 487 S. E. 2d, at 199.

 Unlike the Virginia Supreme Court, the state trial judge omitted any reference to Lockhart and simply relied on our opinion in Strickland as stating the correct standard for judging ineffective-assistance claims. With respect to the prejudice component, he wrote:

 “Even if a Petitioner shows that counsel’s performance was deficient, however, he must also show prejudice. Petitioner must show ‘that there is a reasonable probability that but for counsel’s unprofessional errors, the result … would have been different.’ Strickland, 466 U. S. at 694. ‘A reasonable probability is a probability sufficient to undermine confidence in the outcome.’ Id. Indeed, it is insufficient to show only that the errors had some conceivable effect on the outcome of the proceeding, because virtually every act or omission of counsel would meet that test. Id. at 693. The petitioner bears the ‘highly demanding’ and ‘heavy burden’ in establishing actual prejudice.” App. 417.

The trial judge analyzed the ineffective-assistance claim under the correct standard; the Virginia Supreme Court did not.

 We are likewise persuaded that the Virginia trial judge correctly applied both components of that standard to Williams’ ineffectiveness claim. Although he concluded that counsel competently handled the guilt phase of the trial, he found that their representation during the sentencing phase fell short of professional standards—a judgment barely disputed by the State in its brief to this Court. The record establishes that counsel did not begin to prepare for that phase of the proceeding until a week before the trial. Id., at 207, 227. They failed to conduct an investigation that would have uncovered extensive records graphically describing Williams’ nightmarish childhood, not because of any strategic calculation but because they incorrectly thought that state law barred access to such records. Had they done so, the jury would have learned that Williams’ parents had been imprisoned for the criminal neglect of Williams and his siblings,²⁰ that Williams had been severely and repeatedly beaten by his father, that he had been committed to the custody of the social services bureau for two years during his parents’ incarceration (including one stint in an abusive foster home), and then, after his parents were released from prison, had been returned to his parents’ custody.

 Counsel failed to introduce available evidence that Williams was “borderline mentally retarded” and did not advance beyond sixth grade in school. Id., at 595. They failed to seek prison records recording Williams’ commendations for helping to crack a prison drug ring and for returning a guard’s missing wallet, or the testimony of prison officials who described Williams as among the inmates “least likely to act in a violent, dangerous or provocative way.” Id., at 569, 588. Counsel failed even to return the phone call of a certified public accountant who had offered to testify that he had visited Williams frequently when Williams was incarcerated as part of a prison ministry program, that Williams “seemed to thrive in a more regimented and structured environment,” and that Williams was proud of the carpentry degree he earned while in prison. Id., at 563–566.

 Of course, not all of the additional evidence was favorable to Williams. The juvenile records revealed that he had been thrice committed to the juvenile system—for aiding and abetting larceny when he was 11 years old, for pulling a false fire alarm when he was 12, and for breaking and entering when he was 15. Id., at 534–536. But as the Federal District Court correctly observed, the failure to introduce the comparatively voluminous amount of evidence that did speak in Williams’ favor was not justified by a tactical decision to focus on Williams’ voluntary confession. Whether or not those omissions were sufficiently prejudicial to have affected the outcome of sentencing, they clearly demonstrate that trial counsel did not fulfill their obligation to conduct a thorough investigation of the defendant’s background. See 1 ABA Standards for Criminal Justice 4–4.1, commentary, p. 4–55 (2d ed. 1980).

 We are also persuaded, unlike the Virginia Supreme Court, that counsel’s unprofessional service prejudiced Williams within the meaning of Strickland. After hearing the additional evidence developed in the postconviction proceedings, the very judge who presided at Williams’ trial and who once determined that the death penalty was “just” and “appropriate,” concluded that there existed “a reasonable probability that the result of the sentencing phase would have been different” if the jury had heard that evidence. App. 429. We do not agree with the Virginia Supreme Court that Judge Ingram’s conclusion should be discounted because he apparently adopted “a per se approach to the prejudice element” that placed undue “emphasis on mere outcome determination.” 254 Va., at 26–27, 487 S. E. 2d, at 200. Judge Ingram did stress the importance of mitigation evidence in making his “outcome determination,” but it is clear that his predictive judgment rested on his assessment of the totality of the omitted evidence rather than on the notion that a single item of omitted evidence, no matter how trivial, would require a new hearing.

 The Virginia Supreme Court’s own analysis of prejudice reaching the contrary conclusion was thus unreasonable in at least two respects. First, as we have already explained, the State Supreme Court mischaracterized at best the appropriate rule, made clear by this Court in Strickland, for determining whether counsel’s assistance was effective within the meaning of the Constitution. While it may also have conducted an “outcome determinative” analysis of its own, 254 Va., at 27, 487 S. E. 2d, at 200, it is evident to us that the court’s decision turned on its erroneous view that a “mere” difference in outcome is not sufficient to establish constitutionally ineffective assistance of counsel. See supra, at 26. Its analysis in this respect was thus not only “contrary to,” but also, inasmuch as the Virginia Supreme Court relied on the inapplicable exception recognized in Lockhart, an “unreasonable application of” the clear law as established by this Court.

 Second, the State Supreme Court’s prejudice determination was unreasonable insofar as it failed to evaluate the totality of the available mitigation evidence—both that adduced at trial, and the evidence adduced in the habeas proceeding—in reweighing it against the evidence in aggravation. See Clemons v. Mississippi, 494 U. S. 738, 751–752 (1990). This error is apparent in its consideration of the additional mitigation evidence developed in the postconviction proceedings. The court correctly found that as to “the factual part of the mixed question,” there was “really … n[o] … dispute” that available mitigation evidence was not presented at trial. 254 Va., at 24, 487 S. E. 2d, at 198. As to the prejudice determination comprising the “legal part” of its analysis, id., at 23–25, 487 S. E. 2d, at 198–199, it correctly emphasized the strength of the prosecution evidence supporting the future dangerousness aggravating circumstance.

 But the state court failed even to mention the sole argument in mitigation that trial counsel did advance— Williams turned himself in, alerting police to a crime they otherwise would never have discovered, expressing remorse for his actions, and cooperating with the police after that. While this, coupled with the prison records and guard testimony, may not have overcome a finding of future dangerousness, the graphic description of Williams’ childhood, filled with abuse and privation, or the reality that he was “borderline mentally retarded,” might well have influenced the jury’s appraisal of his moral culpability. See Boyde v. California, 494 U. S. 370, 387 (1990). The circumstances recited in his several confessions are consistent with the view that in each case his violent behavior was a compulsive reaction rather than the product of cold-blooded premeditation. Mitigating evidence unrelated to dangerousness may alter the jury’s selection of penalty, even if it does not undermine or rebut the prosecution’s death-eligibility case. The Virginia Supreme Court did not entertain that possibility. It thus failed to accord appropriate weight to the body of mitigation evidence available to trial counsel.

V

 In our judgment, the state trial judge was correct both in his recognition of the established legal standard for determining counsel’s effectiveness, and in his conclusion that the entire postconviction record, viewed as a whole and cumulative of mitigation evidence presented originally, raised “a reasonable probability that the result of the sentencing proceeding would have been different” if competent counsel had presented and explained the significance of all the available evidence. It follows that the Virginia Supreme Court rendered a “decision that was contrary to, or involved an unreasonable application of, clearly established Federal law.” Williams’ constitutional right to the effective assistance of counsel as defined in Strickland v. Washington, 466 U. S. 668 (1984), was violated.

 Accordingly, the judgment of the Court of Appeals is reversed, and the case is remanded for further proceedings consistent with this opinion.

It is so ordered.