This is an appeal from a judgment of the District Court for the Northern District of California denying petitioners’ writ of habe-as corpus. The two petitioners, Gibson and Justice, are presently serving life sentences in California state prisons for the 1973 murder conviction of a prison guard at San *852Quentin. Gibson and Justice allege that the trial jury that convicted them impermissi-bly relied upon facts that were not in evidence. In doing so they claim they were denied their constitutional right to a fair trial. The district court, which reviewed the state court record, but did not take additional evidence, concluded that any error that may have occurred was harmless beyond a reasonable doubt. Our review of the historical facts in this case convinces us that there is a reasonable possibility that the jury’s consideration of facts not introduced into evidence contributed to the jury’s verdict. Accordingly, we reverse the judgment.
On July 21, 1971, Leo Davis, an unarmed guard at the San Quentin prison hospital, was stabbed to death. The state’s case linking Gibson and Justice to the crime was based principally upon the testimony of three witnesses and the introduction of certain physical evidence.
The state’s chief witness was Herman Johnson who testified that he was an eyewitness to the killing. On July 19, 1971, Johnson claimed he was stabbed in the neck by Gibson and Justice shortly after he had been transferred to San Quentin from Sole-dad. The attack was allegedly provoked by Johnson’s willingness to give a statement implicating other inmates in a crime committed at Soledad. After the attack, Johnson was taken to the prison hospital where he was placed in a room on the second floor. Davis was stationed in the hallway outside his door.
According to Johnson, at around noon on July 21 he caught a glimpse of Gibson through a window in the door to his room and heard scuffling. He testified that upon going to the door he was able to see Gibson and Justice stab Davis while a third man held him.
The defendants impeached Johnson’s testimony with the testimony of other inmates who stated that they had witnessed the July 19 attack on Johnson, and Gibson and Justice were not among the attackers. The defense implied that there was a possibility that Johnson’s testimony was biased because a decision on a parole violation committed by Johnson was allegedly held in abeyance until after his testimony at the trial. The defense also introduced evidence indicating that Johnson had a poor reputation for veracity.
Johnson’s eyewitness account was corroborated by two other prosecution witnesses. Another inmate with the last name of Johnson, Charles Johnson, testified that he had seen Gibson, Justice and a third man in a shower room on the second floor of the hospital within a short time of when the murder occurred.
There was also testimony from Ivan Kranzelic who shared a hospital ward on the fourth floor with Gibson. Gibson had gained admittance to the hospital after allegedly suffering an injury in football practice the day before the murder. Kranzelic testified that on the morning of July 21 he had seen an inmate give a prison-made knife to Gibson, who in turn gave it to Justice. Kranzelic overheard Gibson make various statements such as they would have to do “it” that day. Around noon, Gibson, Justice, and a third inmate left the ward and returned about twenty minutes later. Kranzelic testified that Justice’s hand was bleeding when he returned and, as he sat on Gibson’s bed, he used a roll of toilet paper to try to stop the bleeding. At the time these events were alleged to occur Kranzelic was recovering from foot surgery and had been given several doses of morphine.
At trial the state introduced evidence showing that the murder victim had blood type “0.” Justice had blood type “AB.” It was further shown that the blood stains found on the murder weapon, on Gibson’s bed where Justice had sat, and on the roll of toilet paper, which was also recovered, all were of blood type “AB.”
The trial lasted from December 1972 until April 1973. The jury deliberated for two and a half days before returning a guilty verdict on April 18, 1973. A month later, on May 17, 1973, Gibson and Justice made a motion for a new trial on the grounds that the jury had impermissibly relied upon evidence not produced in court. Affidavits *853submitted by the jurors reveal two incidents of misconduct.
At one point during the deliberations, Juror Colin Grist went to an encyclopedia to confirm his belief that blood type “AB” was rare. He reported his findings back to other members of the jury. Jurors Chapman, Gauger and Cox remembered some comment by Grist. Grist was unsure whether he made the comment before or after the balloting on Justice.
Another juror, Mona Gauger, stated that while the jury was considering Kranzelic’s testimony she had commented on the dosage of morphine. Other jurors were unsure what effect the morphine may have had on Kranzelic, which Gauger interpreted as a request for more information. She then consulted a medical encyclopedia from which she apparently concluded that the morphine dosage was too small to have affected Kranzelic’s perceptions. At least eight other jurors remembered some comment by Gauger, but there was some disagreement as to exactly what she had said.
In both cases there were apparently some comments in the jury room to the effect that the jury should not consider this evidence.
The trial judge denied the motion for a new trial. In his oral ruling the judge considered the applicable test to be that of People v. Watson, 46 Cal.2d 818, 299 P.2d 243, 254 (Cal.1956), cert. denied 355 U.S. 846, 78 S.Ct. 70, 2 L.Ed.2d 55 (1957), which, in the courts of California, requires reversal for an error when it is “reasonably probable that a result more favorable to the appealing party would have been reached in the absence of the error.” (Emphasis added.) The judge specifically declined to apply the stricter test of Chapman v. California, 386 U.S. 18, 24, 87 S.Ct. 824, 828, 17 L.Ed.2d 705 (1967), which requires, in the case of federal constitutional errors, that the error be harmless “beyond a reasonable doubt.”
In his ruling, the judge twice indicated that were he to apply the Chapman standard, he would have granted the motion:
I will state for the record that I cannot say that it [the jury misconduct] was harmless beyond a reasonable doubt.
... I don’t think the Chapman case applies to this situation. If it does and I were to apply it, I would feel, I think, compelled to reach a different result.
The trial judge’s decision was affirmed by an unpublished opinion of the California Court of Appeal. Petitions for certiorari to the California Supreme Court and the United States Supreme Court were denied. In denying the petitioners’ subsequent writ of habeas corpus, the federal district court judge did not specifically decide whether the jury misconduct in this case amounted to constitutional error, but concluded that if there were constitutional error it was harmless beyond a reasonable doubt.
We first must decide whether the California trial judge applied the appropriate test in determining the effect of the impermissi-bly considered evidence. In United States v. Vasquez, 597 F.2d 192 (9th Cir. 1979), we defined the test to be applied when a jury acquires evidence that has not been introduced into the record.
[T]he appellant is entitled to a new trial if there existed a reasonable possibility that the extrinsic material could have affected the verdict.
597 F.2d at 193 (emphasis added). The “reasonable possibility” test of Vasquez is equivalent in severity to the harmless error rule applicable to constitutional errors under Chapman.1
Vasquez involved a direct appeal from a trial in a federal district court. Therefore, although it is intimated by the opinion, there was no need to decide whether the *854“reasonable possibility” test is compelled by constitutional considerations that would make it applicable to the collateral review of a state court judgment.2 We now conclude that the Vasquez test is compelled by constitutional considerations.
In Turner v. Louisiana, 379 U.S. 466, 85 S.Ct. 546, 13 L.Ed.2d 424 (1965), the Court refers to sixth amendment rights in considering the impact of outside influence on a jury:
In the constitutional sense, trial by jury in a criminal case necessarily implies at the very least that the “evidence developed” against a defendant shall come from the witness in a public courtroom where there is full judicial protection of the defendant’s right of confrontation, of cross-examination, and of counsel.
379 U.S. at 472-73, 85 S.Ct. at 549-50.3
Turner involved the possibility of prejudice created where the prosecution’s leading witnesses were the same two deputy sheriffs who watched over the jury during the time it was sequestered for Turner’s trial. Although the case did not involve the introduction of extra-record facts, the implication of the statement is that it would apply in such a case.4 A year after Turner, the court reversed a state court decision where a court bailiff made various prejudicial statements about a defendant to jurors.
The court noted the applicability of the confrontation clause:
[T]he statements of the bailiff to the jurors are controlled by the command of the Sixth Amendment . . .
Parker v. Gladden, 385 U.S. 363, 364, 87 S.Ct. 468, 470, 17 L.Ed.2d 420 (1966):
The Second Circuit has granted habeas corpus relief in two cases where extrinsic material may have influenced a state court verdict.5 Opinions from other circuits have also cited sixth amendment violations when a jury has considered facts outside of those which have been introduced at trial.6
These decisions have noted that when a jury considers facts that have not been introduced in evidence, a defendant has effectively lost the rights of confrontation, cross-examination, and the assistance of counsel with regard to jury consideration of the extraneous evidence. In one sense the violation may be more serious than where these rights are denied at some other stage of the proceedings because the defendant may have no idea what new evidence has been considered. It is impossible to offer evidence to rebut it, to offer a curative instruction, to discuss its significance in argument to the jury, or to take other tactical steps that might ameliorate its impact.7 We believe that the California *855trial judge erred in applying a reasonable probability standard and that the proper standard to be applied is whether it can be concluded beyond a reasonable doubt that extrinsic evidence did not contribute to the verdict.
We must next review the record to determine the effect, if any, of the extrinsic evidence considered by the jury in this case.
The state’s case against Gibson and Justice is a strong one. Nevertheless, if the jury had readily accepted Johnson’s eyewitness testimony it seems unlikely that they would have deliberated for so long to reach a verdict.8 Furthermore, the fact that at least two jurors believed that it was necessary to obtain more evidence is, by itself, an indication that there may have been a need to resolve some lingering hesitation or uncertainty.
The extra-record evidence in this case is of some import in bolstering Johnson’s testimony. Evidence that blood type “AB” is rare would logically strengthen the inference that it had been Justice who had possessed the murder weapon and not some unknown assailant with the same type of blood. The trial judge had ruled that evidence concerning the rarity of “AB” blood was inadmissible. The prosecution took an immediate appeal from this ruling, which is some indication of the importance the state attached to this evidence.9 As to Kranzelic’s testimony, if the jury concluded that his perceptions were not affected by morphine this would certainly strengthen his credibility-
The trial judge candidly acknowledged that were he to apply the test for constitutional error that we have concluded is required in this case, he would have granted the motion for a new trial. Although we are not bound by this evaluation of the historical facts, see e. g., Brown v. Allen, 344 U.S. 443, 456, 73 S.Ct. 397, 406, 97 L.Ed. 469 (1953); Developments in the Law-Federal Habeas Corpus, 83 Harv.L.Rev. 1038, 1113 (1970), we nevertheless believe that under the circumstances the state trial judge’s conclusion should be given a great deal of weight. The state judge presided over Gibson and Justice’s four-month trial and observed the demeanor of the witnesses. At the time of his ruling he was in a far better position to evaluate the credibility of the witnesses than we are now, reviewing a cold record eight years after the trial.
We conclude that there is a reasonable possibility that the extrinsic evidence affected the verdict. Accordingly, the petitioners’ writ of habeas corpus must be granted unless the State of California elects to grant a new trial within sixty days.
REVERSED.
. The court noted in Chapman:
There is little, if any difference between .. . “whether there is a reasonable possibility that the evidence complained of might have contributed to the conviction” and requiring the beneficiary of a constitutional error to prove beyond a reasonable doubt that the error complained of did not contribute to the verdict obtained."
386 U.S. at 24, 87 S.Ct. at 826 (emphasis added).
. Such a test might conceivably be based upon the supervisory powers of the court. See Murphy v. Florida 421 U.S. 794, 797-98, 95 S.Ct. 2031, 2034-35, 44 L.Ed.2d 589 (1975).
. The sixth amendment to the constitution provides in part that an “accused shall enjoy the right ... to be confronted with the witnesses against him . .. and to have the Assistance of Counsel for his defense.”
. In several cases the court has reversed state court convictions without a showing that specific extra-record facts reached a jury during its deliberations. The cases have involved situations where there has been such pervasive pre-trial publicity that it could be presumed that the jury’s verdict was based on either pre-judgment of the defendant or consideration of extraneous facts. See Sheppard v. Maxwell, 384 U.S. 333, 86 S.Ct. 1507, 16 L.Ed.2d 600 (1966); Estes v. Texas, 381 U.S. 532, 85 S.Ct. 1628, 14 L.Ed.2d 543 (1965); Irvin v. Dowd, 366 U.S. 717, 81 S.Ct. 1639, 6 L.Ed.2d 751 (1961).
. Bulger v. McClay, 575 F.2d 407, 411 (2d Cir. 1978), cert. denied 439 U.S. 915, 99 S.Ct. 290, 58 L.Ed.2d 263 (1978); Owen v. McMann, 435 F.2d 813 (2d Cir. 1970), cert. denied 402 U.S. 906, 91 S.Ct. 1373, 28 L.Ed.2d 646 (1971). See also Downey v. Peyton, 451 F.2d 236 (4th Cir. 1971).
. Government of Virgin Islands v. Gereau, 523 F.2d 140, 150 (3rd Cir. 1975) cert. denied 424 U.S. 917, 96 S.Ct. 1119, 47 L.Ed.2d 323 (1976) (“consideration by the jury of extra-record facts about the case ... [is] prima facie incompatible with the Sixth Amendment.”); United States v. Thomas, 463 F.2d 1061, 1063 (7th Cir. 1972) ; Farese v. United States, 428 F.2d 178, 179 (5th Cir. 1970).
. The Fifth Circuit has referred metaphorically to extraneous evidence as “the dagger of hidden evidence.” United States v. Howard, 506 F.2d 865, 866 (5th Cir. 1975).
. The length of jury deliberations has been cited as a factor of some importance in Parker v. Gladden, 385 U.S. 363, 365, 87 S.Ct. 468, 470, 17 L.Ed.2d 420 (1966) (26 hours) and Dallago v. United States, 427 F.2d 546, 559 (D.C.Cir.1969) (5 days). In opposing the petitioner’s motion for a new trial the prosecution stated that the jury actually only deliberated for “possibly nine hours.” Presumably the prosecution meant that this nine hour period was spread over two and one half days. Even accepting the prosecution’s version, however, it does not seem possible that the jury would have deliberated nine hours over several days if the jurors did not have serious questions as to the credibility of the eyewitnesses.
. The fact that a jury has acquired inadmissible evidence has been cited in United States v. Vasquez, 597 F.2d 192, 194 (9th Cir. 1979), and Dallago v. United States, 427 F.2d 546, 558 (D.C.Cir.1969).