with whom Justice Brennan joins and with whom Justice Blackmun joins as to Part II, concurring in part and concurring in the judgment.
I
I agree with the Court that the psychiatric examination on which Dr. Grigson testified at the capital sentencing proceeding was in bald violation of Estelle v. Smith, 451 U. S. 454 *261(1981), and that petitioner’s death sentence should be vacated. I write separately because I believe the Court errs in applying harmless-error analysis to this Sixth Amendment violation. It is my view that the unique nature of a capital sentencing determination should cause this Court to be especially hesitant ever to sanction harmless-error review of constitutional errors that taint capital sentencing proceedings, and even if certain constitutional errors might properly be subject to such harmless-error analysis, a violation of Estelle v. Smith is not such an error.
Until today’s ruling, this Court never had applied harmless-error analysis to constitutional violations that taint the sentencing phase of a capital trial. In deciding to apply harmless-error analysis to the Sixth Amendment violation in this case, I believe the Court fails to adequately consider the unique nature of a capital sentencing proceeding and a sentencer’s decision whether a defendant should live or die. The Court’s analysis is also flawed in that it fails to accord any noticeable weight to the qualitative difference of death from all other punishments.
Unlike the determination of guilt or innocence, which turns largely on an evaluation of objective facts, the question whether death is the appropriate sentence requires a profoundly moral evaluation of the defendant’s character and crime. See California v. Brown, 479 U. S. 538, 545 (1987) (O’Connor, J., concurring) (a death sentence should “reflect a reasoned moral response to the defendant’s background, character, and crime”); Enmund v. Florida, 458 U. S. 782, 801 (1982) (capital defendant’s “punishment must be tailored to his personal responsibility and moral guilt”). Moreover, although much of the Court’s capital jurisprudence since Furman v. Georgia, 408 U. S. 238 (1972), has been focused on guiding and channeling the decision whether death is the appropriate sentence in a specific case, the sentencer nonetheless is afforded substantial discretion. See, e. g., McCleskey v. Kemp, 481 U. S. 279, 304-306 (1987); Woodson v. *262North Carolina, 428 U. S. 280 (1976). Even in the face of overwhelming aggravating evidence, the sentencer has discretion to act with leniency and refuse to impose the death sentence. See McCleskey, supra, at 311 (“[Discretionary exercises of leniency [by the sentencer] are final and unre viewable”).
Because of the moral character of a capital sentencing determination and the substantial discretion placed in the hands of the sentencer, predicting the reaction of a sentencer to a proceeding untainted by constitutional error on the basis of a cold record is a dangerously speculative enterprise. As the Court recognized in Caldwell v. Mississippi, 472 U. S. 320, 330 (1986), “[wjhatever intangibles a jury might consider in its sentencing determination, few can be gleaned from an appellate record.” In the same vein, an appellate court is ill equipped to evaluate the effect of a constitutional error on a sentencing determination. Such sentencing judgments, even when guided and channeled, are inherently subjective, and the weight a sentencer gives an instruction or a significant piece of evidence that is later determined to violate a defendant’s constitutional rights is nowhere apparent in the record. In McCleskey v. Kemp, supra, the Court acknowledged that “[i]ndividual jurors bring to their deliberations ‘qualities of human nature and varieties of human experience, the range of which is unknown and perhaps unknowable,’” and their collective judgment of the appropriate sentence is marked by an “inherent lack of predictability.” Id., at 311, quoting Peters v. Kiff, 407 U. S. 493, 503 (1972) (opinion of Marshall, J.). The threat of an erroneous harmless-error determination thus looms much larger in the capital sentencing context than elsewhere.
That threat is of particular concern because of the unique nature of the death sentence. The awesome severity of a sentence of death makes it qualitatively different from all other sanctions. See, e. g., Lockett v. Ohio, 438 U. S. 586, 605 (1978) (plurality opinion). For this reason, the Court has *263emphasized the greater need for reliability in capital cases, and has required that “capital proceedings be policed at all stages by an especially vigilant concern for procedural fairness and for the accuracy of factfinding.” Strickland v. Washington, 466 U. S. 668, 704 (1984) (Brennan, J., concurring in part and dissenting in part); see California v. Ramos, 463 U. S. 992, 998-999 (1983) (“[T]he qualitative difference of death from all other punishments requires a correspondingly greater degree of scrutiny of the capital sentencing determination”). Because of this heightened concern for reliability, “[t]ime and again the Court has condemned procedures in capital cases that might be completely acceptable in an ordinary case.” Barefoot v. Estelle, 463 U. S. 880, 913 (1983) (Marshall, J., dissenting). Harmless-error analysis impinges directly on the reliability of the capital sentencing decision by allowing a court to substitute its judgment of what the sentencer would have done in the absence of constitutional error for an actual judgment of the sentencer untainted by constitutional error.
I therefore have serious doubts whether a constitutional error that infects the sentencing phase of a capital case ever may be considered harmless beyond a reasonable doubt. But even if I could agree that harmless-error analysis is appropriate for certain constitutional errors at the sentencing phase, such a situation is not presented when the error is a violation of the Sixth Amendment under Estelle v. Smith.
hH h — 1
As an initial matter, the Court in Estelle v. Smith gave no hint that harmless-error analysis ever could apply to the admission of psychiatric testimony in a capital sentencing proceeding which was based on an examination of the defendant conducted in violation of his Sixth Amendment right to counsel. After finding constitutional error, the Court simply vacated the death sentence. See 451 U. S., at 473. The failure of the Court to engage in harmless-error analysis *264in Smith is understandable, because the factors on which this Court traditionally has focused to determine whether harmless-error review is appropriate make clear that an Estelle v. Smith violation that taints a capital sentencing proceeding should lead to automatic reversal. First, the potential for actual prejudice resulting from such a violation of Smith is so high that a “case-by-case inquiry into prejudice is not worth the cost.” Strickland v. Washington, supra, at 692. As evidenced in this case, psychiatric testimony is generally of critical importance to the sentencing determination, covering issues of rehabilitative potential, future dangerousness, and individual culpability.1 Moreover, psychiatric testimony on these issues is clothed with a scientific authority that often carries great weight with lay juries. Cf. Ake v. Oklahoma, 470 U. S. 68, 79 (1985) (recognizing “pivotal role” psychiatry has come to play in criminal proceedings).
Second, it is difficult, if not impossible, to accurately measure the degree of prejudice arising from the failure to notify defense counsel of an impending psychiatric examination and the subsequent admission at the sentencing phase of evidence acquired from the examination. Cf. Hamilton v. Alabama, 368 U. S. 52, 55 (1961) (rejecting harmless-error analysis where “the degree of prejudice can never be known”); Holloway v. Arkansas, 435 U. S. 475, 490-491 (1978), (“[A]n inquiry into a claim of harmless error [in a case involving defense counsel’s conflict of interests] would require, unlike *265most cases, unguided speculation”). As I discussed above, the decision whether a defendant should live or die is a discretionary, moral judgment involving a balancing of often intangible factors. Divining the effect of psychiatric testimony on a sentencer’s determination whether death is an appropriate sentence is thus more in the province of soothsayers than appellate judges. In addition, contrary to the Court’s claim, see ante, at 257, the prejudice arising from an Estelle v. Smith violation is not limited to the illegal admission of psychiatric testimony. If defense counsel is properly notified under Smith of the State’s intention to perform a psychiatric examination, the course of subsequent proceedings may be altered significantly. For instance, defense counsel might extensively prepare his client for the examination, or perhaps advise his client to refuse to participate in the examination by the particular psychiatrist; defense counsel also might urge that a different psychiatrist perform the examination. Cf. Estelle v. Smith, 451 U. S., at 471 (defendant “was denied the assistance of his attorneys in making the significant decision of whether to submit to the examination and to what end the psychiatrist’s findings could be employed”). I therefore believe that any attempt to predict the effect of such an Estelle v. Smith violation would require the appellate court to engage in unguided speculation. The confluence of these factors — the likelihood of prejudice and the difficulty in evaluating the degree of that prejudice — together with the heightened concern for reliability in capital cases, convinces me that a psychiatric examination conducted in violation of Estelle v. Smith, and the later admission at a capital sentencing proceeding of psychiatric testimony based on this examination, may never be considered harmless error.2
*266I would have thought that this Court’s decision in Holloway v. Arkansas, supra, already had settled the question whether an Estelle v. Smith violation in a capital case can ever be harmless error. In Holloway we stated: “ ‘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.’. . . 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.” 435 U. S., at 488-489, quoting Glasser v. United States, 315 U. S. 60, 76 (1942). We stated in Estelle v. Smith, supra, that a pretrial examination by a state psychiatrist of a capital defendant is a “critical stage” in a capital case. Id., at 470. As the Court recognized in that case, “the decision to be made regarding the proposed psychiatric evaluation is ‘literally a life or death matter’ and is ‘difficult . . . even for an attorney’ because it requires ‘a knowledge of what other evidence is available, of the particular psychiatrist’s biases and predilections, [and] of possible alternative strategies at the sentencing hearing.’” Id., at 471, quoting Smith v. Estelle, 602 F. 2d 694, 708 (CA5 1979).
The Court attempts to distinguish Holloway by arguing that in that case the “deprivation of the right to counsel affected — and contaminated — the entire criminal proceeding.” Ante, at 257. But Holloway anticipated automatic reversal not only when the deprivation affected the entire proceeding, but also when the deprivation occurred during a “critical stage in, at least, the prosecution of a capital offense.” 435 U. S., at 489 (emphasis added). By focusing on whether the error occurred in a capital case, Holloway exhibited an apprecia*267tion of the heightened concern for reliability in this context— something I believe today’s decision fails to recognize.3
In the end, the Court principally relies on its belief “that a reviewing court can make an intelligent judgment about whether the erroneous admission of psychiatric testimony might have affected a capital sentencing jury. ” Ante, at 258. I do not possess the same confidence in an appellate court’s ability to divine the prejudice arising from such a significant error in a capital sentencing proceeding. In my view, the speculation engendered by harmless-error review of a violation of Estelle v. Smith in the context of a capital sentencing proceeding presents an intolerable danger that the death sentence will be administered erroneously. Accordingly, I do not join in that aspect of the Court’s opinion sanctioning harmless-error analysis for violations of Estelle v. Smith.
The likelihood of actual prejudice arising from the illegal admission of psychiatric testimony is even greater in the context of this case. The Texas capital sentencing statute provides that, in the absence of evidence of provocation by the victim, the court “shall sentence the defendant to death” if the jury finds that the murder was “committed deliberately and with the reasonable expectation that the death of the deceased . . . would result,” and that “there is a probability that the defendant would commit criminal acts of violence that would constitute a continuing threat to society.” Tex. Code Crim. Proc. Ann., Art. 37.071 (Vernon Supp. 1988). The psychiatrist’s evaluation of future dangerousness thus purports to answer one of two questions posed by the statute.
It is also important to note that a violation of petitioner’s Sixth Amendment right to counsel under Estelle v. Smith is easy to identify and, “for that reason and because the prosecution is directly responsible, easy for the government to prevent.” Strickland v. Washington, 466 U. S. 668, 692 (1984). Because the error is in the control of the State and is easy to *266prevent, holding that such a violation will result in automatic reversal does not pose a significant burden on the State.
Moreover, in the present case the Court is unable to cite a single capital case since our decision in Furman v. Georgia, 408 U. S. 238 (1972), in which we have ignored Holloway’s reasoning and have applied harmless-error analysis to a Sixth Amendment violation occurring during a critical stage of the proceedings. The Court cites dicta in Buchanan v. Kentucky, 483 U. S. 402, 425, n. 21 (1987), as an indication of the Court’s willingness to apply harmless-error analysis to the admission of psychological testimony in violation of Estelle v. Smith, 451 U. S. 454 (1981). But the petitioner in Buchanan was not prosecuted for a capital offense, and thus the Court’s indication in that case that harmless-error analysis might apply to the illegal admission of psychological testimony has little relevance in the present context.