dissenting.

As federal judges, we’re subject to a peculiar myopia: We tend to see the federal way of doing things as the best way, often as the be-all and end-all of due process. This preference for the familiar is understandable enough, but it carries with it special dangers when we review state convictions for constitutional error. There are many ways of giving a defendant a fair trial. We sap our federal system of vitality and creativity when we strike down state procedures because they differ from our own.

This is a case in point. Not everyone agrees on how to deal with a jury that indicates it’s deadlocked. The jury shouldn’t keep deliberating if unanimity is beyond reach, but it ought not to be discharged if there’s a reasonable possibility of reaching agreement. To resolve this tension, most federal courts, as well as some state courts, give an Allen charge — a fairly coercive instruction that tells jurors to go back and try again. The Allen charge has been subject to much criticism, however, and; many courts have either modified or rejected it, the California Supreme Court among.them. 1 People v. Gainer, 19 Cal.3d 835 , 139 Cal.Rptr. 861 , 566 P.2d 997 (1977), held that the Allen charge is inherently coercive and thus reversible error.. In its place, the California Supreme Court approved a neutral inquiry into the jury’s numerical division as “an important tool in ascertaining the probability of agreement.” People v. Rodriguez, 42 Cal.3d 730 , 776 n. 14, 230 Cal-Rptr. 667, 726 P.2d 113 (1986). It has repeatedly refused to reconsider this practice. See, e.g., People v. Price, 1 Cal.4th 324, 467 , 3 Cal.Rptr.2d 106 , 821 P.2d 610 (1991); People v. Proctor, 4 Cal.4th 499 , 539 n. 7, 15 Cal.Rptr.2d 340 , 842 P.2d 1100 (1992).

Confronted with a possible jury deadlock, the state trial judge here did precisely what his state supreme court- told him to: He asked the foreperson about the jury’s division. Upon learning there’d been movement in one direction, he exercised the discretion entrusted to him and sent the jury back for further deliberations. Later, when the jury again declared itself deadlocked, the judge again followed his supreme court’s instructions and asked the foreperson for a division. Learning that the jury had moved even closer to agreement, the trial judge again exercised his judgment and sent them back to deliberate. He turned out to be right: The jury returned a verdict later that day. ‘

By overturning the conviction so obtained, my colleagues are casting serious doubt on a procedure the California Supreme Court and a number of other state supreme courts have approved for dealing with apparent jury deadlock. My colleagues hold that the practice of asking for a jury’s division is so inherently coercive it renders a trial fundamentally unfair when combined with even the trivial comments the trial judge made here. This ruling bodes ill for countless convictions in *1480 California and other states that follow the California practice. See p. 1481 infra.

My colleagues give no persuasive reason for such heavy-handed interference with the power of state supreme courts to resolve difficult issues of trial practice. Today’s ruling moves us one step farther away from a truly federal system where the state courts serve as laboratories for shaping the flexible and evolving standards of due process. It moves us one step closer to a world where our way of doing things becomes the only way, where state courts dare not deviate from our procedures lest they risk wholesale invalidation of their convictions.

A. First things first. The majority overturns the conviction here because it finds the trial judge’s actions unduly coercive. We should pause to ask whether the coercion of a few holdout jurors in a state criminal trial can ever violate due process. Jury coercion violates due process in a federal criminal trial because the defendant there is constitutionally entitled to a unanimous jury. Johnson v. Louisiana, 406 U.S. 356, 371 , 92 S.Ct. 1620, 1638 , 32 L.Ed.2d 152 (1972) (Powell, J., concurring). Coercion of holdout jurors impairs the federal defendant’s Sixth Amendment right to a fair trial because it casts doubt on whether the jury was unanimous. But the defendant in a state criminal trial has no constitutional right to a unanimous jury; an 11-1, 10-2 or even 9-3 guilty verdict is consistent with due process. Id. at 363 , 92 S.Ct. at 1625 (majority opinion). If a state can dispense with the concurrence of one, two, three — perhaps even more-r-holdout jurors altogether, why can’t it try to secure the agreement of those same jurors through moral suasion? 2 The latter procedure seems an a fortiori case of the former.

Nor is jury coercion inherently prejudicial to the defense. Our perception of this may be distorted because every time we see jury coercion, the jury has convicted; it’s easy to conclude that coercion always harms defendants. But pressuring holdout jurors is just as likely to lead to an acquittal as to a conviction, depending on the direction the jury is leaning. When judicial pressure is applied without regard to whether the jury favors a conviction or an acquittal, it’s hard to say this denies the defendant a fair trial.

Finally, it might be argued that if a defendant is entitled to a unanimous jury as a matter of state law, it violates due process to let the judge coerce the holdout jurors. But it all turns on how state law defines unanimity. Some states view unanimity in the same pristine fashion as the federal courts; others may temper the concept, allowing some pushing and tugging to get there. It surely doesn’t violate due process for state law to défine unanimity more flexibly than federal law.

The Supreme Court has, nevertheless, suggested that jury coercion in a state trial can violate due process. See Lowenfield v. Phelps, 484 U.S. 231, 241 , 108 S.Ct. 546, 552 , 98 L.Ed.2d 568 (1988). Absent more specific guidance it’s probably safest to assume it does. But in deciding how bad jury coercion is — and how much leeway to allow states in dealing with apparently hung juries — we must keep in mind the teachings of cases like Johnson, Williams v. Florida, 399 U.S. 78 , 90 S.Ct. 1893 , 26 L.Ed.2d 446 (1970) (12-person jury is not a constitutional requirement), and Cupp v. Naughten, 414 U.S. 141 (1973) (jury instruction universally condemned in federal courts is not unconstitutional), which guarantee states a good deal of latitude in conducting jury trials.

B. Even if we assume jury coercion violates due process, we must next ask whether what happened here was coercive. Ever since the Supreme Court’s ipse dixit in Brasfield v. United States, 272 U.S. 448 (1926), the federal courts have taken it as an article of faith that a judicial inquiry into the jury’s division is coercive. But Brasfield didn’t explain why it found this practice coercive nor is the reason immediately apparent. After all, asking the jury how it’s divided carries *1481 with it no explicit message; it exhorts the jurors to do nothing at all. I suppose the inquiry could be understood as denoting some judicial impatience, perhaps some veiled suggestion that jurors in the minority give in to their peers. Bmsfield’s determination that holdout jurors will perceive and be swayed by such a hidden message is less law than psychology, more a guess than a fact. One could conclude just as easily that the inquiry is not coercive because the jury will read nothing .into it.

A number of states take this view. California is but one of seven states to have determined that a judicial inquiry into the jury’s division is noncoercive and entirely appropriate. See State v. Keaulana, 784 P.2d 328 (Haw.1989); State v. Jones, 641 P.2d 708 (Wash.1982); State v. Morris, 476 S.W.2d 485 (Mo.1971); Dunford v. State, 614 P.2d 1115 (Okla.Crim.App.1980); Sharplin v. State, 330 So.2d 591 (Miss.1976); State v. Rickerson, 625 P.2d 1183 (N.M.1981). To this group, two other states can be added that also seem to approve the practice: See Odom v. State, 682 S.W.2d 445, 448 (Tex.Ct.App.1984) (defendant “has not cited a Texas ease, nor have we found one, which makes the mere inquiry as to the numerical division of the jury reversible error per se”); State v. Schamburge, 344 So.2d 997, 1001 (La.1977) (“The trial judge did no more than make a general inquiry as to prospects for a verdict and reminded the jury that ten of the twelve had to agree upon a verdict.”).

Another group of states — seven by my count — condemns the practice summarily. See Gumer v. Commonwealth, No. 85-CA-1923-MR (Ky.Ct.App.April 25, 1986); Graham v. State, 325 Md. 398 , 601 A.2d 131 (1992); Hyman Reiver & Co. v. Rose, 51 Del. 397 , 147 A.2d 500 (1958); People v. Wilson, 390 Mich. 689 , 213 N.W.2d 193 (1973); Rodriguez v. State, 559 So.2d 678 (Fla.Ct.App.1990); Kersey v. State, 525 S.W.2d 139 (Tenn.1975); State v. George, 219 Mont. 377 , 711 P.2d 1379 (1986). They rely on Brasfield or a closely analogous rationale.

An additional eleven states have addressed the issue but do not have a clear-cut position. Among them are states like North Carolina, which find that “[s]ueh inquiries are not inherently coercive....” State v. Bussey, 321 N.C. 92 , 361 S.E.2d 564, 567 (1987). States like Illinois and Arizona disapprove the inquiry but consider it harmless. People v. Craddock, 163 Ill.App.3d 1039 , 115 Ill.Dec. 1, 6 , 516 N.E.2d 1357, 1362 (1987); People v. Kirk, 76 Ill.App.3d 459 , 31 Ill.Dec. 835, 841 , 394 N.E.2d 1212, 1218 (1979); State v. Roberts, 131 Ariz. 513 , 642 P.2d 858, 861 (1982). A substantial number mention, but make no ádverse comment on, the trial judge’s inquiry into the jury’s division, and focus instead on supplemental Allertr-type instructions. See State v. Marsh, 260 Or. 416 , 490 P.2d 491, 501 (1971); Hardin v. State, 225 Ark. 602 , 284 S.W.2d 111, 115 (1955); Black v. State, 255 Ga. 668 , 341 S.E.2d 436 (1986); State v. Anderson, 247 Minn. 469 , 78 N.W.2d 320, 326-27 (1956). Some states adopt a balancing approach, considering the inquiry as part of a broader question of coercion. Richardson v. State, 508 So.2d 289, 293-94 (Ala.Crim.App.1987); People v. Austin, 185 Colo. 229 , 523 P.2d 989, 993-94 (1974); State v. Cornell, 266 N.W.2d 15, 19 (Iowa 1978); White v. State, 95 Nev. 881 , 603 P.2d 1063, 1065 (1979).

In determining whether a particular state court procedure violates due process,-it’s highly relevant whether that state is alone in its view. While a deviation from the norm doesn’t automatically offend the Constitution, failure to follow the collected wisdom of other American jurisdictions can raise doubts about whether the procedure is consistent with due process. 3 The Supreme -Court’s analysis in Beck v. Alabama, 447 U.S. 625 , 100 S.Ct. 2382 , 65 L.Ed.2d 392 (1980), illustrates this point. Beck held that the defendant in a capital trial is entitled, as a matter of due process, to a lesser included offense instruction. The Court reasoned that “the nearly universal acceptance of the rule in both state and federal courts [as a matter of statutory or common law] establishes the *1482 value to the defendant of this procedural safeguard.” Id. at 637 , 100 S.Ct. at 2389 . There is no such universal agreement that a judicial inquiry into a jury’s division is coercive. The majority’s assumption to the contrary — which lies at the heart of its ruling— is without support.

C. Even assuming the judge’s inquiry into the jury’s division and his related comments were coercive, we must next ask how coercive. My colleagues recognize that “coercion must be more severe to constitute a violation of due process than to invoke our supervisory powers,” Maj. op. at 1477, but they then stand this admonition on its head. The cautionary part of the Allen charge — the absence of which the majority sees as the great failing here — is designed to offset the coercive part of the charge. Thus, in order to vacate a state conviction for failure to give a cautionary reminder, the judge’s actions here had to be more coercive than the dynamite part of the Allen charge.

By any standard, what happened here — if it was coercive at all — was much less coercive than the typical Allen charge, where the trial judge looks the holdout jurors in the eye and says:

United States v. Ajiboye, 961 F.2d 892 , 893 n. 1 (9th Cir.1992).

Contrast what happened here: The trial judge asked the jury to tell him how badly they were split and whether there’d been any movement. His comments were relatively mild, more by way of explanation than exhortation. After the first inquiry, when he learned the jury was moving toward unanimity, the judge said: “Well, that’s what’s important to me because of the nature of the type of case. I want to find out that there has been movement.” His comment after the second inquiry was entirely neutral: “All right. Due to the fact we have had that type of movement, I would request, then, to finish the rest of today and see where we are' at that point in time. All right? Okay.”

What happened here simply is not on a par with — let alone more coercive than — the dynamite portion of the Allen charge. The majority errs even according to the standard it has set for itself.

D. Even were we to adopt the majority’s approach today, we would have no business applying it retroactively to petitioner or to any other state prisoner whose conviction has become final. The Supreme Court in Teague v. Lane, 489 U.S. 288 , 109 S.Ct. 1060 , 103 L.Ed.2d 334 (1989) (plurality opinion), held that where the federal courts adopt a new rule as a concomitant of due process, the rule must — except in rare cases — be applied only prospectively. 4 The reasons for this are amply demonstrated by the case before us: What the trial judge .did here wasn’t an aberration; he followed a practice the California Supreme Court had approved repeatedly. Giving today’s rule retroactive application will have a profound and unsettling effect on hundreds of convictions entered by the courts of California and other states in this circuit. In many cases, there will be no hope of retrying the defendants. Retroactive application of the new rule thus significantly impairs these states’ legitimate interests in maintaining convictions secured through procedures their courts reasonably believed conformed with due process.

*1483 That the rule the majority adopts is new can hardly be doubted. No federal court— no- court anywhere — has held that a state judge’s inquiry into a jury’s numerical division is sufficiently coercive to require an Allen-type cautionary instruction. The only ease the majority cites, Jones v. Norvell, 472 F.2d 1185 (6th Cir.1973), involved not an inquiry as to division alone but, as my colleagues acknowledge, an instruction that “it was [the jury’s] duty to reach, a verdict.” Maj. op. at 1478. Reliance on Jones is particularly misplaced in light of the Sixth Circuit’s own limited reading of it in Williams v. Parke, 741 F.2d 847 (6th Cir.1984). The Williams court confronted a situation far more egregious than that here: a state trial court that had not only inquired into the jury’s division, but given a truncated Allen charge — one that was all coercion and no cautionary reminder. Id. at 849, 851 ; see also id. at 853 (Contie, J., dissenting). The court distinguished Jones on the ground that, “in Jones, the trial judge determined not only the jury breakdown, but also the majority’s inclination.” Id. at 851. In addition, “the judge in Jones ... wrongly asserted that the twelve people of the jury ‘are the only ones’ who could reach a verdict in the ease, thus appearing to assert the impossibility of retrial.” Id. at 851-52.

Sixth Circuit case law can more fairly be characterized as conflicting with the majority’s opinion than supporting it. If the inquiry as to division plus the coercive portion of the Allen charge didn’t violate due process in Williams ,, the trial judge’s neutral inquiries and oblique comments here surely did not.

Other federal circuits have uniformly rejected the kind of claim the majority sustains here. Ellis v. Reed, 596 F.2d 1195 (4th Cir.1979), dealt with a situation where the judge had asked about the jury’s division, and was told “eleven to one.” He then gave a fairly strong version of the Allen charge, reminding the jurors that if they could not agree “the time of the Court will again have to be consumed in the trial of this action;” that “it is your duty to try to reconcile your differences and to reach a verdict;” that “a mistrial will mean that another jury will have to be selected to hear this case and the evidence again; and it’s long and complicated;” and finally that “I want to emphasize the fact that it is your duty to do whatever you can to reason this matter over as reasonable men and women and attempt to reconcile your differences_” Id. at 1196 . The jury reached agreement in eight minutes. Id. By any standard, the coercion in Ellis was far more severe than anything that happened here.

Although the state trial judge did include a brief cautionary instruction, the approach,of Ellis surely diverges from that of the majority here. Ellis , grounded its analysis on the Supreme Court’s rulings in Johnson, Williams and Cupp and concluded that these are matters that should be left to the state courts:

Id. at 1200. Nothing in the Fourth Circuit’s opinion paves the way for our ruling today.

■The Seventh Circuit took a similarly narrow view of the federal prerogatives in this area. In United States ex rel. Kirk v. Director, Dept. of Corrections, 678 F.2d 723 (7th Cir.1982), the court confronted the petitioner who’d been the defendant in People v. Kirk. See p. 1481 supra. In Kirk , the Illinois Court of Appeals held that the inquiry into jury division was error but harmless. The Seventh Circuit characterized the case as involving an inquiry into jury division followed by a modified Allen charge, although neither its opinion nor that of the Illinois appellate court discloses the exact nature of the charge. 5 The Seventh Circuit *1484 rejected the argument that an inquiry as to jury division was inherently coercive; the court didn’t.focus on whether the trial judge reminded the jurors of their duty not bo give up honestly held beliefs. Instead, the court held that the inquiry had to be considered in light of all the circumstances of the case, and that the defendant had shown nothing to suggest the inquiry was coercive. Kirk surely does not foreshadow the result here for purposes of Teague .

The Eighth Circuit dealt with the issue in Cornell v. State, 628 F.2d 1044 (8th Cir.1980), reviewing the ruling of the state court in State v. Cornell, 266 N.W.2d 15 (Iowa 1978). See p. 1481 supra. In Cornell, the trial judge asked how the jury was divided after more than 15 hours of deliberation, and was told 7 to 5. At that point, the trial court gave a lengthy Allen-type charge that can best be described as 95% dynamite and 5% caution. 6 The Eighth Circuit denied relief, noting that “neither the inquiry nor the Allen charge, nor the two in combination, was coercive of the jury’s ultimate verdict of guilty.” Cornell, 628 F.2d at 1048 . Cornell, like Ellis , is difficult to square with our ruling today and is arguably in direct conflict. It certainly in no way presages our decision for purposes of Teague .

Finally, there’s our prior ruling in Locks v. Sumner, 703 F.2d 403 (9th Cir.1983), which involved an inquiry as to numerical division, apparently not followed by any comment from the trial judge. Id.- at 405 n. 1. We upheld the conviction; citing Cornell, Ellis and Cupp , we concluded:

Id. at 407. If Locks is not actually in conflict with our opinion today, it surely contains no suggestion, no hint, that an Alleiu-type cautionary reminder is required whenever the judge asks the jury how it’s divided. 7

The long and short of it is that today’s decision is out of step "with those of other federal courts. It’s the first ease anywhere that equates a neutral inquiry into the jury’s division with the dynamite portion of an Allen charge; it is the first federal court to reverse a state conviction under these circumstances. If Teague does not bar retroactive application in these circumstances, it has no meaning at all.

E. The Supreme Court has repeatedly directed the federal courts to stick to their limited role in reviewing state convictions. *1485 See, e.g., Estelle v. McGuire, — U.S.-, -, 112 S.Ct. 475, 480 , 116 L.Ed.2d 385 (1991) (“[W]e reemphasize that it is not the province of a federal habeas court to reexamine state court determinations on state law questions.”); see also Pulley v. Harris, 465 U.S. 37, 41 , 104 S.Ct. 871, 874-75 , 79 L.Ed.2d 29 (1984). But, as this ease illustrates, many federal judges find it difficult to separate their role of supervising the federal courts from their narrower role of policing possible constitutional violations' in state habeas cases. Nor is this an isolated instance. In two other recent habeas cases, my colleagues casually equated federal rules of procedure with constitutional requirements. See Henry v. Estelle, 993 F.2d 1423 (9th Cir.1993); McKinney v. Rees, 993 F.2d 1378 (9th, Cir.1993). Henry is a particularly egregious example, containing, as it does, the following passage: “This inference is impermissible. Fed.R.Evid. 404.” Henry, 993 F.2d at 1427 . I fear a pattern is emerging where federal judges create constitutional rules that allow them to remake state procedure in the federal image.

As Chief Justice Burger once noted, “[w]e should be cautious about constitutionalizing every procedural device found useful in federal courts, thereby foreclosing the States from experimentation with different approaches which are equally compatible with constitutional principles.” Crist v. Bretz, 437 U.S. 28, 39 , 98 S.Ct. 2156, 2163 , 57 L.Ed.2d 24 (1978) (Burger, C.J., dissenting). If the current trend continues, the concept of due process will disappear altogether, to be replaced by the Federal Rules of Criminal Procedure. See Williams, 399 U.S. at 136-38 , 90 S.Ct. at 1925-26 (Harlan, J., concurring in result). This is not, I respectfully suggest, a wonderful idea.

Conclusion

Dealing with an apparently deadlocked jury is not easy. Reasonable jurists certainly might respond in more than one way. My colleagues today foreclose the State of California from pursuing the approach it prefers merely because this approach doesn’t conform to the mold cast by the federal courts. By applying the decision retroactively, the majority compounds the damage. I dissent.

. Even the federal courts are far from unanimous on how to deal with jury deadlock. While we allow the Allen charge in its original form, see United States v. Seawell, 583 F.2d 416 (9th Cir.1978), the Third Circuit has forbidden it altogether, see United States v. Graham, 758 F.2d 879 (3d Cir.1985).

. The matter might be different if the jury has made it- clear it's leaning towards a conviction; jurors might then interpret a nudge towards unanimity as a hint that the judge believes the defendant is guilty. If the judge "urge[s] his own view of the guilt or innocence of the accused,” he may interfere with the defendant's right to a trial by the jury. Billed v. United States, 184 F.2d 394, 403 (D.C.Cir.1950).

. The converse isn’t necessarily true; a procedure might violate due process even though a substantial number of states follow it. Yet, in determining whether an inquiry into numerical division is coercive, it surely matters that a significant number of state supreme courts have considered the procedure and held it isn’t.

. There are two exceptions to this general rule: if the rule places "certain kinds of primary, private individual conduct beyond the power of the criminal lawmaking authority to proscribe;" or if it requires the observance of "those procedures that ... are 'implicit in the concept of ordered liberty.’ ” 489 U.S. at 307 , 109 S.Ct. at 1073 (citations omitted). It goes Without saying that neither applies here.

. People v. Kirk says that the trial judge gave a Prim, instruction. In People v. Prim, 53 Ill.2d 62 , *1484 289 N.E.2d 601 (1972), the Illinois Supreme Court approved the use of the ABA recommended instruction in lieu of the standard Allen charge. The ABA instruction contains a milder form of the Alien dynamite charge, as well as a cautionary reminder. Id.

. The only portion of the instruction which can be deemed cautionary is the following phrase: "[W]hile this verdict must be the conclusion of each juror and not a mere acquiescence of the jurors in order to reach agreement....” Cornell, 628 F.2d at 1046 . This was sandwiched between a variety of exhortations such as: "A proper regard for the judgment of others will greatly aid us in forming our own judgment,” and "there is no reason to think a jury better qualified could ever be chosen,” and "if the members of the jury differ in their views of the evidence, such difference of opinion should cause them all to scrutinize the evidence more closely and to reexamine the grounds of their opinion,” and "your duty is to decide the issues of fact which have been submitted to you, if you can conscientiously do so,” and "you should lay aside all mere pride of opinion and should bear in mind that the jury room is no place for espousing and maintaining in a spirit of controversy.”

. Also relevant is United States v. Akbar, 698 F.2d 378 (9th Cir.1983), cited in Lodes. In Akbar , the district judge conducted an unusual jury poll to determine which jurors wanted certain testimony reread. During the course of that procedure, the judge learned not only that the jury was split 11 to 1, but also the identity of the holdout juror. There was no cautionary instruction, yet the court reluctantly affirmed. See also id. at 380-81 (Kennedy, J., concurring).