with whom Justice Blackmun joins, dissenting.
Concerned that respondent Fretwell would otherwise receive the “windfall” of life imprisonment, see ante, at 366, 370, the Court today reaches the astonishing conclusion that deficient performance by counsel does not prejudice a defendant even when it results in the erroneous imposition of a death sentence. The Court’s aversion to windfalls seems to disappear, however, when the State is the favored recipient. For the end result in this case is that the State, through the coincidence of inadequate representation and fortuitous timing, may carry out a death sentence that was invalid when imposed.
*377This extraordinary result rests entirely on the retrospective application of two changes in the law occurring after respondent’s trial and sentencing. The first of these changes, on which the Court relies explicitly, affected the eligibility of defendants like Fretwell for the death penalty. The second change, never directly identified as such, is the Court’s unprincipled transformation of the standards governing ineffective-assistance claims, through the introduction of an element of hindsight that has no place in our Sixth Amendment jurisprudence.
In my view, the Court of Appeals correctly determined that “fundamental unfairness exists when a prisoner receives a death sentence rather than life imprisonment solely because of his attorney’s error.”1 The Court’s post hoc rationale for avoiding this conclusion, self-evident until today, is both unconvincing and unjust.
I
“Unless a defendant charged with a serious offense has counsel able to invoke the procedural and substantive safeguards that distinguish our system of justice, a serious risk of injustice infects the trial itself.” Cuyler v. Sullivan, 446 U. S. 335, 343 (1980). For that reason, we have held squarely that the right to counsel guaranteed by the Constitution is a right to the “effective assistance of counsel.” See United States v. Cronic, 466 U. S. 648, 654 (1984). Absent competent counsel, ready and able to subject the prosecution’s case to the “crucible of meaningful adversarial testing,” there can be no guarantee that the adversarial system will function properly to produce just and reliable results. Id., at 656. See Strickland v. Washington, 466 U. S. 668, 684-687 (1984).
In some cases, the circumstances surrounding a defendant’s representation so strongly suggest abridgment of the *378right to effective assistance that prejudice is presumed. When, for instance, counsel is prevented from offering assistance during a critical phase of the proceedings,2 or labors under a conflict of interest that affects her performance,3 then we assume a breakdown in the adversarial process that renders the resulting verdict unreliable. See United States v. Cronic, 466 U. S., at 658-660. We need not, even if we could, inquire further into the precise nature of the prejudice sustained. See Glasser v. United States, 315 U. S. 60, 75-76 (1942). It is enough that the adversarial testing envisioned by the Sixth Amendment has been thwarted; the result is constitutionally unacceptable, and reversal is automatic. See United States v. Cronic, 466 U. S., at 656-657; Holloway v. Arkansas, 435 U. S. 475, 489 (1978).4
In Strickland v. Washington, 466 U. S. 668 (1984), the Court decided that certain errors by counsel will give rise to a similar presumption of adversarial breakdown. Because the consequences that attend such a presumption — the setting aside of a conviction or sentence — are so serious, the Court took pains to limit the class of errors that would support an ineffective-assistance claim. First, an error must be so egregious that it indicates “deficient performance” by counsel, falling outside the “wide range of reasonable profes*379sional assistance.” Id., at 687, 689. Second, the error must be so severe that it gives rise to prejudice, defined quite clearly in Strickland as “a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.” Id., at 694. Many significant errors, as the Court recognized in Kimmelman v. Morrison, 477 U. S. 365, 381-382 (1986), will not meet this “highly demanding” standard. But those that do will require reversal, not because they deprive a defendant of some discrete and independent trial right, but because, as Strickland held, they reflect performance by counsel that has “so undermined the proper functioning of the adversarial process that the trial cannot be relied on as having produced a just result.” 466 U. S., at 686.
Under this well-established standard, as the District Court and Court of Appeals both determined, respondent is entitled to relief on his ineffective-assistance claim. That his counsel’s performance was so wanting that it was “deficient” for Strickland purposes is not contested. Nor can it be seriously disputed that the decision reached would “reasonably likely have been different,” id., at 696, but for counsel’s failure to make a double-counting objection supported by Eighth Circuit law.5 Under Strickland, this is the end of the inquiry. Respondent has identified an error of such magnitude that it falls within the narrow class of attorney errors precluding reliance on the outcome of the proceeding. See id., at 691-692. In Sixth Amendment terms, it is as though respondent had shown an actual conflict of interest, or the complete absence of counsel during some part of the sentencing proceeding: The adversary process has malfunctioned, and the resulting verdict is therefore, and without more, constitutionally unacceptable.
*380This is not, however, the standard that the Court applies today. Instead, the Court now demands that respondent point to some additional indicia of unreliability, some specific way in which the breakdown of the adversarial process affected respondent’s discrete trial rights. Ante, at 369-370. But this is precisely the kind of harmless-error inquiry that the Court has rejected, time and again, in the Sixth Amendment context. When a criminal proceeding “loses its character as a confrontation between adversaries,” United States v. Cronic, 466 U. S., at 656-657, the harm done a defendant is as certain as it is difficult to define. Accordingly, we consistently have declined to require that a defendant who faces the State without adequate assistance show how he is harmed as a result. See Cuyler v. Sullivan, 446 U. S., at 349; Holloway v. Arkansas, 435 U. S., at 489-491; Hamilton v. Alabama, 368 U. S. 52, 55 (1961); Williams v. Kaiser, 323 U. S. 471, 475-477 (1945). “The right to have the assistance of counsel is too fundamental and absolute to allow courts to indulge in nice calculations as to the amount of prejudice arising from its denial.” Glasser v. United States, 315 U. S., at 76.6
*381The Court compounds its error by insisting that respondent make his newly required showing from the vantage point of hindsight. Hindsight has no place in a Sixth Amendment jurisprudence that focuses, quite rightly, on protecting the adversarial balance at trial. Respondent was denied “the assistance necessary to justify reliance on the outcome of the proceeding,” Strickland v. Washington, 466 U. S., at 692, because his counsel’s performance was so far below professional standards that it satisfied Strickland’s first prong, and so severely lacking that the verdict “would reasonably likely have been different absent the errors,” id., at 696, under the second prong. It is simply irrelevant that we can now say, with hindsight, that had counsel failed to make a double-counting objection four years after the fact, his performance would have been neither deficient nor prejudicial. For as it happened, counsel’s failure to object came at a time when it signified a breakdown in the adversarial process. A post hoc vision of what would have been the case years later has no bearing on the force of this showing.
Not surprisingly, the Court’s reliance on hindsight finds no support in Strickland itself. Strickland makes clear that the merits of an ineffective-assistance claim must be “viewed as of the time of counsel’s conduct.” Id., at 690. As the Court notes, this point is stated explicitly with respect to Strickland’s first prong, the quality of counsel’s performance. Ante, at 371-372. What the Court ignores, however, is that the same point is implicit in Strickland’s entire discussion of the second prong. By defining prejudice in terms of the effect of counsel’s errors on the outcome of the proceedings, *382based on the “totality of the evidence before the judge or jury,” 466 U. S., at 695, the Strickland Court establishes its point of reference firmly at the time of trial or sentencing.
To justify its revision of the Strickland standards forjudg-ing ineffective-assistance claims, the Court relies in large part on Nix v. Whiteside, 475 U. S. 157 (1986). Ante, at 370. Nix cannot, however, perform the heavy duty the Court assigns it. A rather unusual case, Nix involved a claim that counsel was ineffective because he refused to present a defense based on perjured testimony. It should suffice to say here that reliance on perjured testimony and reliance on current Court of Appeals case law are not remotely comparable, and that to suggest otherwise is simply disingenuous. But if further distinction is needed, we need not search far to find it.
First, the Court’s decision in Nix rests in part on the conclusion that counsel’s refusal to cooperate in presentation of perjury falls “well within ... the range of reasonable professional conduct acceptable under Strickland.” Nix v. Whiteside, 475 U. S., at 171; cf. United States v. Cronic, 466 U. S., at 656, n. 19 (“Of course, the Sixth Amendment does not require that counsel do what is impossible or unethical. If there is no bona fide defense to the charge, counsel cannot create one . . .”). In other words, ineffective-assistance claims predicated on failure to make wholly frivolous or unethical arguments will generally be dispensed with under Strickland’s first prong, without recourse to the second, and hence will not raise the questions at issue in this case.
To the extent that Nix does address Strickland’s second, or “prejudice,” prong, it does so in a context quite different from that presented here. In Strickland, the Court cautioned that assessment of the likelihood of a different outcome should exclude the possibility of “a lawless decision-maker,” who fails to “reasonably, conscientiously, and impartially appl[y] the standards that govern the decision.” 466 U. S., at 695. The Nix Court faced what is perhaps a *383paradigmatic example of the “lawlessness” to which Strickland referred, in the suggestion that perjured testimony might have undermined the decisionmaker’s judgment, and concluded quite correctly that the defendant could not rely on any outcome-determinative effects of perjury to make his claim. Nix v. Whiteside, 475 U. S., at 175; see also id., at 186 (Blackmun, J., concurring in judgment). I do not read the Court’s decision today as suggesting that a state trial court need fear the label “lawless” if it follows the decision of a United States Court of Appeals on a matter of federal constitutional law. Accordingly, Nix’s discussion of perjury and lawlessness is simply inapposite to the issues presented here.
II
It is not disputed in this case that the performance of respondent’s counsel was so deficient that it met the Strickland standard. What deserves emphasis here is the proven connection between that deficiency and the outcome of respondent’s sentencing proceeding, as well as the presumptive effect of counsel’s performance on the adversarial process itself.
Respondent was convicted of committing murder in the course of a robbery. The Arkansas trial court then held a separate sentencing hearing, devoted exclusively to the question whether respondent was eligible for the death penalty, or would instead receive a life sentence without parole. The State relied on two aggravating circumstances to establish its right to execute respondent. The first — the alleged purpose of avoiding arrest — was found by the jury to be unsupported by the evidence. The second — that the felony was committed for purposes of pecuniary gain — was obviously supported by the evidence, as respondent had already been convicted of robbery in connection with the murder. Thus, the critical question on which respondent’s death eligibility turned was whether it was permissible, as a matter of law, to “double count” by relying on pecuniary gain as an *384aggravating circumstance and also on robbery as an element of the crime.
Counsel’s duty at this stage of the proceedings was clear. In addition to general investigation and preparation for the penalty phase, counsel’s primary obligation was to advise the trial judge about the correct answer to this crucial question of law. Had he handled this professional responsibility with anything approaching the “reasonableness” demanded by Strickland, 466 U. S., at 687-691, he would have found an Eighth Circuit case directly in point, addressing the same Arkansas statute under which respondent was sentenced and holding such double counting unconstitutional. Collins v. Lockhart, 754 F. 2d 258, 261-265, cert. denied, 474 U. S. 1013 (1985). The failure to find that critically important case constitutes irrefutable evidence of counsel’s inadequate performance. The fact that Collins was later overruled does not minimize in the slightest the force of that evidence.
Moreover, had counsel made a Collins objection to the pecuniary gain aggravating circumstance, we must assume that the trial court would have sustained it. As the District Court stated: “Although Collins has since been overruled, it was the law in the Eighth Circuit at the time of [respondent’s] trial and this Court has no reason to believe that the trial court would have chosen to disregard it.” 739 F. Supp. 1334, 1337 (ED Ark. 1990). Neither petitioner nor the Court relies on disagreement with this finding. See n. 5, swpra. Nor could they. As we explained in Strickland, it is not open to the State to argue that an idiosyncratic state trial judge might have refused to follow circuit precedent and overruled a Collins objection. 466 U. S., at 695.
Applying Strickland to these facts, the District Court correctly held that counsel’s failure to call the trial judge’s at: tention to Collins constituted ineffective assistance and “seriously undermined the proper functioning of the adversarial process.” 739 F Supp., at 1336. Because it granted relief on this basis, the District Court found it unnecessary to *385reach additional ineffective-assistance claims predicated on counsel’s alleged failure to investigate or prepare for the penalty phase. Id., at 1337-1338.7 By the time the case reached the Court of Appeals, deficient performance was conceded, and the Eighth Circuit had only to affirm the District Court conclusion that “a reasonable state trial court would have sustained an objection based on Collins had Fretwell’s attorney made one.” 946 F. 2d, at 577.8
Thus, counsel’s deficient performance, in the form of his failure to discover Collins and bring it to the court’s attention, is directly linked to the outcome of respondent’s sentencing proceeding. Because of counsel’s error, respondent received the death penalty rather than life imprisonment. *386946 F. 2d, at 577. Under Strickland, of course, respondent need not show quite so much; it is sufficient that “the decision reached would reasonably likely have been different absent the errors.” 466 U. S., at 696. A fortiori, a showing of outcome determination as strong as that made here is enough to support a Strickland claim.
In my judgment, respondent might well be entitled to relief even if he could not show prejudice as defined by Strickland’s second prong. The fact that counsel’s performance constituted an abject failure to address the most important legal question at issue in his client’s death penalty hearing gives rise, without more, to a powerful presumption of breakdown in the entire adversarial system. That presumption is at least as strong, if not stronger, than the inferences of adversarial malfunction that required reversal in cases like Holloway and Glasser, see supra, at 377-378. In other words, there may be exceptional cases in which counsel’s performance falls so grievously far below acceptable standards under Strickland’s first prong that it functions as the equivalent of an actual conflict of interest, generating a presumption of prejudice and automatic reversal. I think this may well be one of those cases in which, as we wrote in Holloway, reversal would be appropriate “even if no particular prejudice is shown and even if the defendant was clearly guilty.” 435 U. S., at 489 (internal quotation marks and citation omitted).
Of course, we need not go nearly so far to resolve the case before us. Under the Strickland standard that prevailed until today, respondent is entitled to relief on his ineffective-assistance claim, having shown both deficient performance and a reasonable likelihood of a different outcome. The Court can avoid this result only by effecting a dramatic change in that standard, and then applying it retroactively to respondent’s case. In my view, the Court’s decision marks a startling and most unwise departure from our commitment *387to a system that ensures fairness and reliability by subjecting the prosecution’s case to meaningful adversarial testing.
m
Changes in the law are characteristic of constitutional adjudication. Prior to 1985, most of those changes were in the direction of increasing the protection afforded an individual accused of crime. To vindicate the legitimate reliance interests of state law enforcement authorities, however, and in recognition of the state interest in preserving the outcome of trials adhering to contemporaneous standards, the Court often refused to apply its new rules retroactively.9 In Teague v. Lane, 489 U. S. 288, 310 (1989), the Court gave full expression to its general policy of allowing States “to keep in prison defendants whose trials and appeals conformed to then-existing constitutional standards,” holding that the claims of federal habeas petitioners will, in all but exceptional cases, be judged under the standards prevailing at the time of trial.10
Since 1985, relevant changes in the law often have been in a different direction, affording less rather than more protec*388tion to individual defendants.11 An evenhanded approach to retroactivity would seem to require that we continue to evaluate defendants’ claims under the law as it stood at the time of trial. If, under Teague, a defendant may not take advantage of subsequent changes in the law when they are favorable to him, then there is no self-evident reason why a State should be able to take advantage of subsequent changes in the law when they are adverse to his interests.
The Court, however, takes a directly contrary approach here. Today’s decision rests critically on the proposition that respondent’s ineffective-assistance claim is to be judged under the law as it exists today, rather than the law as it existed at the time of trial and sentencing. Ante, at 372. In other words, respondent must make his case under Perry v. Lockhart, 871 F. 2d 1384 (CA8), cert. denied, 493 U. S. 959 (1989), decided four years after his sentencing; unlike the State, he is not entitled to rely on “then-existing constitutional standards,” Teague, 489 U. S., at 310, which rendered him ineligible for the death penalty at the time that sentence was imposed.
I have already explained why the Court’s reliance on hindsight is incompatible with our right to counsel jurisprudence. It is also, in my judgment, inconsistent with case law that insists on contemporaneous constitutional standards as the benchmark against which defendants’ claims are to be measured. A rule that generally precludes defendants from taking advantage of postconviction changes in the law, but allows the State to do so, cannot be reconciled with this *389Court’s duty to administer justice impartially. Elementary fairness dictates that the Court should evaluate respondent’s ineffective-assistance claim under the law as it stood when he was convicted and sentenced — under Collins, and also under Strickland as it was understood until today.
As I see it, the only windfall at issue here is the one conferred upon the State by the Court’s decision. Had respondent’s counsel rendered effective assistance, the State would have been required to justify respondent’s execution under a legal regime that included Collins. It is highly unlikely that it could have met this burden in the Arkansas courts, see supra, at 384-385, and it almost certainly could not have done so in the federal courts on habeas review. Now, however, the State is permitted to exploit the ineffective assistance of respondent’s counsel, and the lapse in time it provided, by capitalizing on postsentencing changes in the law to justify an execution. Because this windfall is one the Sixth Amendment prevents us from bestowing, I respectfully dissent.
946 F. 2d 571, 577 (CA8 1991).
See, e. g., Geders v. United States, 425 U. S. 80 (1976) (attorney-client consultation prevented during overnight recess); Hamilton v. Alabama, 368 U. S. 62 (1961) (assistance denied during arraignment).
See, e. g., Cuyler v. Sullivan, 446 U. S. 335 (1980) (actual conflict adversely affecting performance constitutes reversible error); Glasser v. United States, 316 U. S. 60 (1942) (joint representation of codefendants with inconsistent interests, over objection, constitutes reversible error).
“[T]his Court has concluded that the assistance of counsel is among those constitutional rights so basic to a fair trial that their infraction can never be treated as harmless error. Accordingly, when a defendant is deprived of the presence and assistance of his attorney, either throughout the prosecution or during a critical stage in, at least, the prosecution of a capital offense, reversal is automatic.” Holloway v. Arkansas, 435 U. S., at 489 (internal quotation marks and citation omitted).
Neither petitioner nor the Court today directly challenges the District Court’s unambiguous conclusion that “the trial court would have followed the ruling in Collins had trial counsel made an appropriate motion.” 739 F. Supp. 1334, 1337 (ED Ark. 1990).
It is worth noting that Kimmelman v. Morrison, 477 U. S. 365 (1986), is entirely consistent with this line of case law, rendering petitioner’s reliance on that case misplaced. In Kimmelman, the Court held that although certain Fourth Amendment violations are themselves not cognizable on federal habeas review, see Stone v. Powell, 428 U. S. 465 (1976), counsel’s failure to litigate such Fourth Amendment claims competently may still give rise to a cognizable ineffective-assistance claim. In other words, attorney error gives rise to an ineffective-assistance claim not because it is connected to some other, independent right to which a defendant is entitled, but because in itself it "upset[s] the adversarial balance between defense and prosecution,” so that the trial is rendered unfair and the verdict suspect. 477 U. S., at 374.
That Kimmelman at one point refers to the necessity for a “meritorious” Fourth Amendment claim, id., at 382, as emphasized by Justice O’Connor in her concurrence, ante, at 374, represents no more than straightforward application of Strickland’s outcome-determinative test for *381prejudice. Simply put, an attorney’s failure to make a Fourth Amendment objection will not alter the outcome of a proceeding if the objection is meritless, and hence would not be sustained. Nothing in Kimmelman suggests that failure to make an objection supported by current precedent, and hence likely to be sustained, would amount to anything less than ineffective assistance.
It should come as no surprise that counsel’s conduct gave rise to additional ineffective-assistance claims, founded on other deficiencies. An attorney who makes one error of Strickland proportions is unlikely to have turned in a performance adequate in all other respects. For instance, it may well be more than coincidence that the same counsel who failed to discover United States Court of Appeals precedent holding application of the Arkansas capital sentencing statute to defendants like his client unconstitutional also failed to convince the jury of the existence of any mitigating circumstances in his client’s favor. 739 F. Supp., at 1335. The connection in this case between counsel’s failure to make a Collins objection and his overall preparation and investigation for the penalty phase seems perfectly clear. Nothing in the Court’s opinion today would preclude the District Court, on remand, from considering the lack of an objection as evidence relevant to the larger question of the adequacy of counsel’s penalty phase preparation and investigation.
I cannot agree with the gloss put on the opinion below by the Court, ante, at 368, and by Justice Thomas in his concurrence, ante, at 375. There is nothing in the text of that opinion to suggest that the Court of Appeals believed the Arkansas trial court bound by the Supremacy Clause to obey Eighth Circuit precedent. The Court of Appeals simply noted that the trial court was “bound by the Supremacy Clause to obey federal constitutional law,” 946 F. 2d, at 577 (emphasis added), which is why Eighth Circuit precedent giving content to that law would have been relevant to the trial court’s decisionmaking. I see no reason to infer from its plain and correct statement of the law that the Eighth Circuit actually meant to express the view addressed by Justice Thomas.
See, e. g., Stovall v. Denno, 388 U. S. 293, 300 (1967) (“factors of reliance and burden on the administration of justice” mandate against retroactive application of United States v. Wade, 388 U. S. 218 (1967), and Gilbert v. California, 388 U. S. 263 (1967), establishing right to counsel at pretrial identification); Johnson v. New Jersey, 384 U. S. 719 (1966) (declining to apply Miranda v. Arizona, 384 U. S. 436 (1966), retroactively); Tehan v. United States ex rel. Shott, 382 U. S. 406 (1966) (Griffin v. California, 380 U. S. 609 (1965), prohibiting adverse comment on a defendant’s silence, does not apply retroactively).
See also Engle v. Isaac, 456 U. S. 107, 128-129, n. 33 (1982) (discussing “frustration” of state courts when they “faithfully apply existing constitutional law” only to have change in constitutional standards applied retroactively).
See, e. g., Payne v. Tennessee, 501 U. S. 808 (1991) (Eighth Amendment does not preclude use of victim impact evidence against capital defendant at sentencing; overruling Booth v. Maryland, 482 U. S. 496 (1987), and South Carolina v. Gathers, 490 U. S. 805 (1989)); Arizona v. Fulminante, 499 U. S. 279 (1991) (harmless-error rule applicable to admission of involuntary confessions); Duckworth v. Eagan, 492 U. S. 195 (1989) (Miranda warnings adequate despite suggestion that lawyer will not be appointed until after interrogation); Florida v. Riley, 488 U. S. 445 (1989) (police may search greenhouse from helicopter at altitude of 400 feet without warrant).