The issue in this case is whether the ex post facto clause is violated when a statutory amendment that increases a penalty to be imposed upon the revocation of supervised release is applied in a case in which the underlying offense was committed before the amendment was adopted but the conduct that led to revocation of supervised release occurred afterwards. We conclude that applying the amendment under such circumstances violates the ex post facto clause.

I.

Stuart Jeffrey Paskow conspired to receive the proceeds of a bank robbery and received of the proceeds of a bank robbery, in violation of 18 U.S.C. §§ 371 , 2113(c), in April and May of 1988. He pled guilty and was sen tenced in August, 1990, to eight months’ imprisonment and three years of supervised release.

On December 31, 1988, several months after Paskow committed that underlying offense, an amendment to the supervised release statute, 18 U.S.C. § 3583 , became effective. Anti-Drug Abuse Act of 1988, 102 Stat. 4181 , 4464, Pub.L. 100-690 § 7303 (b). The amendment provides:

18 U.S.C. § 3583 (g). Before the amendment was enacted, violation of supervised release for possession of a controlled substance carried no mandatory minimum penalty. Indeed, before the amendment, the only restriction on the court’s discretion was a requirement that no sentence could exceed two years (if the underlying offense was a Class C or D felony) or three years (if the underlying offense was a Class B felony). 18 U.S.C. § 3583 (e). Thus, before the addition of the amendment, the statute imposed a ceiling on a court’s discretion, but set no floor. The effect of the amendment was to set a floor: one-third of the term of supervised release. Practically speaking, following the amendment, Paskow was subject to a mandatory minimum one-year sentence for violating supervised release. Prior to the amendment, he could have received no sentence at all.

Paskow began the supervised release portion of his sentence in December, 1990. The terms of his release required that he submit to random drug tests. In August, 1992, after random testing revealed that he had used marijuana and cocaine, his supervised release was revoked and he was sentenced to twelve months in prison, one-third of his term of supervised release, the mandatory minimum under the amendment. 18 U.S.C. § 3583 (g).

At his revocation hearing, Paskow objected to his sentence on the ground that the amendment to the supervised release statute, as applied, violated the ex post facto clause. The district court held that, because the behavior for which his supervised release was revoked took place after the amendment became effective, the ex post facto prohibition was not violated. [RT 14-15]. Paskow now appeals. While the district court’s ruling has great appeal intuitively, a close analysis of Supreme Court and other precedent persuades us that we are required to reverse.

II.

Article I of the Constitution provides that neither Congress nor any state shall pass any ex post facto law. U.S. Const, art. I, § 9, cl. 3, art. I, § 10, cl. I. An ex post facto law is not simply one that makes criminal an act that was lawful at the time it was committed, or a law that increases a sentence following the commission of the act for which punishment is imposed. The ex post facto provision applies to a wide range of changes affecting trial procedures and the mechanics of punishment.

The Supreme Court first analyzed the Constitution’s ex post facto clause in Calder v. Bull, 3 Dall. 386 , 1 L.Ed. 648 (1788). There, Justice Chase stated:

Id. at 390 . Justice Chase’s formulation remains the core of the ex post facto prohibition. Miller v. Florida, 482 U.S. 423, 429-30 , 107 S.Ct. 2446, 2450-51 , 96 L.Ed.2d 351 (1987).

Since Colder, a two-pronged test has been developed to determine whether the application of a penal law violates the ex post facto clause. “[Fjirst, the law ‘must be retrospective, that is, it must apply to events occurring before its enactment’; and second, ‘it must disadvantage the offender affected by it.’ ” Miller v. Florida, 482 U.S. at 429 , 107 S.Ct. at 2451 (quoting Weaver v. Graham, 450 U.S. 24, 29 , 101 S.Ct. 960, 964 , 67 L.Ed.2d 17 (1981)). These two factors must be assessed in connection with the date of the defendant’s offense, not of his conviction or sentencing. Watson v. Estelle, 886 F.2d 1093, 1096 (9th Cir.1989) (describing as “the key inquiry” the provisions of the law “at the time the defendant perpetrated the offense”). See also Weaver v. Graham, 450 U.S. at 30 , 101 S.Ct. at 965 (date of inquiry is the date “when the crime was consummated”).

We now consider, first, whether Paskow was disadvantaged by the application of the amended statute, and, next, whether the statute is retrospective.

A.

The government concedes that the amended statute disadvantages Paskow. Indeed, any argument to the contrary would be foreclosed by Lindsey v. Washington, 301 U.S. 397 , 57 S.Ct. 797 , 81 L.Ed. 1182 (1937). The statute at issue there, at the time the defendant committed his crime, punished grand larceny by a sentence of between six months and fifteen years. By the time the defendant was sentenced, however, a change in the law modified the penalty to a flat fifteen-year sentence. The Court found that the retrospective application of the new law violated the ex post facto clause, because the defendant was “deprived of all opportunity to receive a sentence which would give [him] freedom from custody and control prior to the expiration of the 15-year term.” 301 U.S. at 401-02 , 57 S.Ct. at 799 (emphasis added) (quoted with approval, Miller v. Florida, 482 U.S. at 432 , 107 S.Ct. at 2452 ). The principle of Lindsey , then, is that, in determining the disadvantage to a defendant, a court must focus on the change in the defendant’s eligibility to receive a lesser sentence than a new law may permit, regardless of whether the defendant would actually have received the lesser sentence.

Here, before the supervised release statute was amended, Paskow was eligible to receive a sentence of less than one-third the term of his supervised release. 18 U.S.C. § 3583 (e). After the amendment, the district court was no longer free to impose such a sentence. It is the difference between a court’s discretion to sentence a defendant to a particular term — or to one substantially shorter — and the absolute requirement that the defendant be sentenced to at least that term, that implicates the first prong of the ex post facto prohibition. We conclude that the application of the amended statute to Paskow disadvantaged him.

B.

We now consider the more difficult question whether the amended statute is retrospective as applied to Paskow. Following several decades of consistent case law, we conclude that it is.

1.

We find virtually dispositive the strong line of cases that decides this precise issue in connection with revocation of parole (as opposed to the revocation of supervised release at issue here). These cases hold that the ex post facto clause is violated when a parole violator is punished in a way that adversely affects his ultimate release date under a statute that was adopted after the violator committed the underlying offense but before he violated the terms of his parole. For the purposes of an ex post facto analysis, there is absolutely no difference between parole and supervised release. The considerations that underlie each are identical.

The line of cases on which we rely involve parole because, until 1987, federal prisoners were subject to parole rather than to supervised release, see 18 U.S.C. § 3551 et seq. (setting forth the provisions of the Sentencing Reform Act of 1984), and because the states, by and large, still employ the parole system. In these cases, the date of eligibility for parole is tantamount to a presumptive release date. The same is true of a date of eligibility for supervised release. To substitute “supervised release” for “parole” in no way changes the ex post facto analysis. In both cases, the question is at what time the prisoner is to be released from prison. A delay in that date constitutes the same punishment whether it is imposed following a parole violation or a violation of supervised release. Thus, whether the punishment is imposed in connection with the former violation or the latter is irrelevant for ex post facto purposes. If, as is the case, it is an ex post facto violation to impose more severe punishment for a parole violation when the punishment is based on conduct that occurs after the statutory amendment that authorizes the more severe punishment (because the amendment was not in effect at the time the underlying offense was committed), the same is necessarily true in the case of'the imposition of a more severe punishment for a violation of supervised release in like circumstances.

2.

It is clear that, in the case of parole revocation, it is unconstitutional to apply a statute that alters, to the defendant’s disadvantage, the terms under which eligibility for re-parole is calculated, if that statute was enacted after the date of the underlying offense — even though the act upon which revocation is based occurred after the enactment of the statute. The Supreme Court’s affir-mance of the three-judge panel decision in Greenfield v. Scafati 277 F.Supp. 644 (D.Mass.1967), aff'd mem., 390 U.S. 713 , 88 S.Ct. 1409 , 20 L.Ed.2d 250 (1968), makes that proposition indisputable, and necessarily dictates the outcome of this case.

Although the Supreme Court affirmed Greenfield in a memorandum opinion, 390 U.S. 713 , 88 S.Ct. 1409 , 20 L.Ed.2d 250 (1968), its decision is controlling authority. Indeed, the Court has cited its memorandum opinion with approval, Weaver v. Graham, 450 U.S. at 34 , 101 S.Ct. at 966 , and has described it as one of “[t]he Court’s precedents,” 450 U.S. at 37 , 101 S.Ct. at 969 (Blackmum, J., concurring). See also Love v. Fitzharris, 460 F.2d 382, 384 (9th Cir.1972) (citing Greenfield with approval), vacated as moot, 409 U.S. 1100 , 93 S.Ct. 896 , 34 L.Ed.2d 682 (1973).

In Greenfield , a defendant who was incarcerated following revocation of his parole challenged a statute that prohibited any state parole violator from receiving good-conduct credits during his first six months in custody following revocation. 277 F.Supp. at 644-45 . At the time the defendant committed his underlying crime, all prisoners, including parole violators, could accumulate good-conduct credits from the beginning of their incarceration. The new statute, like the amendment to the supervised release statute at issue here, was adopted after the defendant committed his underlying crime, but before he committed the offense for which his parole was revoked. The three-judge court held that application of the statute to the defendant violated the ex post facto clause, because the statute prevented him from being released as early as he might have been had he been permitted to amass good-conduct credits under the statute in effect at the time he committed the underlying crime. Thus, according to the three-judge court and ac cording to the Supreme Court, the statute operated retrospectively and to his detriment. As the three-judge court stated, the effect of the statute was to “extend[] his sentence and increas[e] his punishment” beyond the amount he expected or had notice of when he committed his underlying crime. Id. at 645 (quoting Lembersky v. Parole Bd., 124 N.E.2d 521 , 524, 332 Mass. 290 , 294 (1955)).

Circuit courts that have considered the ex post facto issue have, without exception, followed Greenfield , holding that the ex post facto clause is violated when a defendant’s eligibility for release is adversely affected under a statute that was not in effect at the time of the defendant’s underlying crime but was adopted before the defendant committed the act for which his parole was revoked. Without exception, the circuit courts that have considered the ex post facto issue have looked to the date of the defendant’s original offense, not to the date of the conduct that led to the punishment for the parole violation. Without exception, they have based their holdings on the axiom that, as the Second Circuit phrased it, “parole eligibility is considered an integral part of any sentence.” Shephard v. Taylor, 556 F.2d 648, 654 (2d Cir.1977) (holding that standards in effect at the time of a defendant’s original sentence must be applied at his revocation hearing). See also Fender v. Thompson, 883 F.2d 303, 305 (4th Cir.1989) (discussed below); Schwartz v. Muncy, 834 F.2d 396 , 398 n. 8 (4th Cir.1987) (holding that a defendant’s eligibility for parole for an initial offense could not be adversely altered as a consequence of a second offense, because “parole eligibility is part of the law annexed to the crime at the time of a person’s offense”); Beebe v. Phelps, 650 F.2d 774, 777 (5th Cir.Unit A 1981) (discussed below); Greenfield v. Scafati, 277 F.Supp. at 645 (stating that, had the law authorizing partial forfeiture of eligibility for parole been in effect at the time of the defendant’s original crime, it “would have been contemplated in the sentence, and would therefore be within the maximum term provided for”). Because parole eligibility is part of the sentence for the underlying offense, its terms and conditions are fixed at the moment the underlying offense is complete. Therefore, like the length of a term of incarceration, the conditions affecting parole eligibility cannot be retrospectively altered.

In Beebe v. Phelps, supra, a case factually identical to Greenfield , the Fifth Circuit concluded that the statute at issue, which forfeited good-time credits upon revocation of parole, violated the prohibition on ex post facto laws because it constituted “a sanction that extends the time remaining on petitioner’s original sentence” rather than a punishment for “the second offense” (the parole violation). 650 F.2d at 776 (emphasis added). The Fifth Circuit stated, “The practical effect is a statutory increase in punishment for the first offense, enacted subsequent to the commission of the offense.” Id.

Similarly, the Fourth Circuit, considering a statute that made ineligible for parole any prisoner serving a life sentence who escaped from custody, held that the critical date for ex post facto purposes is not the date of the escape but rather the date of the offense for which the prisoner was originally sentenced. The court stated, “[S]tatutes enacted or amended after a prisoner was sentenced cannot be applied to alter the conditions of or revoke his or her preexisting parole eligibility-notwithstanding that the conduct purportedly triggering application of the statute occurred after its enactment.” Fender v. Thompson, 888 F.2d at 306.

The case law is absolutely clear. Greenfield and its progeny stand for the principle that the date of a prisoner’s eligibility for release on parole is a component of a defendant’s original sentence. Greenfield also compels the conclusion that no adverse action can be taken at a parole revocation hearing that could not have been taken at such a hearing on the day the defendant committed the underlying crime. Finally, it requires us to conclude that the parole revocation must apply no more disadvantageous rules and standards than were in effect at that time. Here, as discussed above, there is no dispute that the amendment to the statute was disadvantageous to Paskow and subjected him to harsher treatment with respect to his release date than did the statute as it existed at the time he committed the underlying offense.

The government makes two arguments in opposition to the application of the Greenfield principle here. It contends that, because Paskow violated the terms of his supervised release after the effective date of the amended statute, he was on notice of the changed penalty for his conduct, and that the amended statute does not increase the penalty for a prior crime, but rather enhances the penalty for the revocation behavior.

These are the identical arguments that were made and rejected in the Greenfield line of cases. See Fender v. Thompson, 883 F.2d at 306-07 . To repeat them here is simply to question the correctness of those decisions. Were we to adopt the government’s arguments, we would be required to jettison twenty-five years of precedent. Because Greenfield constitutes binding Supreme Court authority and because the other circuits that have considered the question have uniformly followed the Greenfield principle, we may not simply decline to apply it here.

3.

We have previously explained generally why supervised release is no different from parole for purposes of the Greenfield line of cases. Nevertheless, some further discussion may be helpful. The government, unfortunately, offers no rational basis for distinguishing the two. Nor can we conceive of any reason at all why one rule should apply in supervised release cases while the opposite rule applies in parole revocation cases.

Supervised release and parole are virtually identical systems. Under each, a defendant serves a portion of a sentence in prison and a portion under supervision outside prison walls. If a defendant violates the terms of his release, he may be incarcerated once more under the terms of his original sentence. More specifically, a defendant’s original sentence determines the length of the term of parole (indirectly) or supervised release (directly). It is also the original sentence that establishes how long the defendant may be required to serve following revocation in the case of both parole and supervised release violations. Finally, it is the original sentence that is executed when the defendant is returned to prison after a violation of the terms of both parole and supervised release.

Revocation of parole is not a punishment for a new offense, although the conduct on which revocation is based may be punished separately. For revocation purposes, the conduct simply triggers the execution of the conditions of the original sentence. Those conditions may not be made more severe, nor may the defendant’s term of rein-carceration after his violation be made more onerous, by any act adopted after he was sentenced. In this respect, there is no conceivable basis for distinguishing between parole and supervised release.

Two courts have specifically considered whether the parole cases apply in the context of supervised release. Both have held that they do, and that the ex post facto clause is violated when the amendment is applied to a defendant who committed his underlying offense before the statute was amended. United States v. Parriett, 974 F.2d 523 (4th Cir.1992), United States v. Flora, 810 F.Supp. 841 (W.D.Ky.1993). Parriett rested its holding on Fender v. Thompson, supra, concluding that no meaningful distinction existed between parole eligibility and supervised release. Therefore, the court’s holding that “the revision of the supervised release statute altered the legal consequences of Parriett’s original crime,” 974 F.2d at 526 & n. 2, naturally followed. In Flora , an exceptionally well-reasoned opinion, the district court concluded that “revocation [of supervised release] and additional imprisonment ] are as much the consequence of the offender’s underlying crime as is the initial term of imprisonment,” and that, therefore, the amended statute may not be applied to a defendant whose underlying offense was committed before its effective date. 810 F.Supp. at 843 . The court arrived at its holding by reasoning from case law, the plain language of the supervised release statute, 18 U.S.C. § 3583 , and the theory of supervised release set forth in Chapter 7 of the sentencing guidelines, United States Sentencing Commission, Guidelines Manual, Ch. 7, Pt. A, § 3(b) (“U.S.S.G. Ch. 7, Pt. A”).

Flora was decided after the government submitted its brief in this case. The government’s only comment on Parriett is to note that the court seemed reluctant to reach its conclusion. The Parriett court stated that the law is absolutely clear in parole cases and that there is simply no way to distinguish supervised release from parole for ex post facto purposes. It found the weight of precedent inescapable. So do we.

It is not only the cases that have considered supervised release and have found no distinction between it and parole, for ex post facto purposes, that compel our conclusion. Our result is also dictated by the text of the supervised release statute and the sentencing guidelines concerning supervised release.

As noted above, the case law makes clear that a defendant’s eligibility for release is “an integral part of any sentence.” Shephard v. Taylor, 556 F.2d at 654 . It is for this reason that statutes that adversely affect such eligibility cannot be applied at a defendant’s revocation hearing if they were adopted after the time of sentencing. The supervised release statute also makes it clear that the terms and conditions of supervised release are a part of the sentence. The statute states, “The court, in imposing a sentence to a term of imprisonment for a felony or a misdemeanor, may include as part of the sentence the requirement that the defendant be placed on a term of supervised release after imprisonment_” 18 U.S.C. § 3583 (a) (emphasis added). Chapter 7 of the sentencing guidelines, which contains the guidelines’ entire discussion of supervised release, states unequivocally that supervised release is “a part of the sentence of imprisonment [imposed] at the time of initial sentencing.” U.S.S.G. Ch. 7, Pt. A, § 2(b). It also states that, in formulating the policy statements relevant to supervised release, the Sentencing Commission explicitly rejected a theory of supervised release that treated violations of conditions “as new federal criminal conduct.” Id. at § 3(b). If a violation is not to be treated as a new substantive offense, it cannot constitute a separate offense for ex post facto purposes.

In sum, the case law and the language of the statute and guidelines cause us to conclude that there is no difference for ex post facto purposes between parole and supervised release. Each is a component of the defendant’s sentence. No conditions affecting either may be changed to his detriment after he commits the underlying crime.

Summary

To summarize our analysis, it is beyond dispute that the ex post facto clause is violated when a parole violator is punished in a way that adversely affects his ultimate release date and the law that authorizes the punishment was adopted after the violator committed the underlying offensé, but before he violated the terms of his parole. The courts have spoken uniformly, and the government’s arguments to the contrary simply repeat contentions that have been rejected over the past twenty-five years.

The issue here is whether the same rule applies to supervised release. Our answer to that question is yes. As the Fourth Circuit stated, there is “no persuasive reason to distinguish between the standards of parole eligibility ... and the conditions for revocation of supervised release.” United States v. Parriett, 974 F.2d at 526 n. 2.

One of the primary reasons for the application of the ex post facto clause is that conditions that affect a defendant’s eligibility for parole are part of his original sentence. So, too, are the conditions that affect the revocation of his supervised release. The language of both the supervised release statute and the sentencing guidelines make this clear. In fact, in formulating the guidelines, the Sentencing Commission expressly rejected the suggestion that a defendant’s behavior in violating the terms of his supervised release should be treated as a new substantive offense.

We believe that the guidelines, the statute, and, especially, the case law, compel our holding. We are bound by an unbroken and consistent line of cases that stretches back over a quarter of a century, including a case affirmed by the Supreme Court, one that the Court itself has declared to be controlling precedent.

In sum, a term of supervised re- ' lease, 18 U.S.C. § 3588 , is “simply part of the whole matrix of punishment which arises out of a defendant’s original crimes.” United States v. Flora, 810 F.Supp. at 842 . Therefore, the ex post facto clause is violated when an amendment to the supervised release statute disadvantages a defendant who committed the underlying offense before the amendment became effective-even if, as here, the defendant committed the act that led to revocation of his supervised release after the amendment was adopted. We therefore hold that, as applied to Paskow, 18 U.S.C. § 3583 (g) violates the cx post facto clause. Accordingly, we vacate his sentence and remand his case for resentencing.

VACATED AND REMANDED.

. Paskow was actually sentenced to a term of fifteen months and 27 days. The additional three months and 27 days were improperly imposed. We resolved the discrepancy between the twelvemonth sentence the judge intended to impose and the sentence actually imposed in an unpublished order. United States v. Paskow, No. 92-50616 (9th Cir. filed Sept. 9, 1993). . “ 'So much importance did the [cjonvention attach to [the ex post facto prohibition], that it is found twice in the Constitution.' " Weaver v. Graham, 450 U.S. 24 , 28 n. 8, 101 S.Ct. 960 , 964 n. 8, 67 L.Ed.2d 17 (1981) (quoting Kring v. Missouri, 107 U.S. 221, 227 , 2 S.Ct. 443, 448 , 27 L.Ed. 506 (1883)). . In addition, the provision in question must be a "law.” See Miller v. Florida, 482 U.S. at 434-35 , 107 S.Ct. at 2453-54 . Here, it is indubitable that the amended statute at issue is a "law” for ex post facto purposes. The government does not argue to the contrary. . Even before the effective date of the Sentencing Reform Act, some offenses required the imposition of a term of supervised release rather than special parole. See, e.g., 21 U.S.C. § 841 (b)(1)(C) (requiring supervised release in addition to imprisonment for specified narcotics offenses). It was pursuant to this statute that the defendant in United States v. Bermudez was sentenced. United States v. Bermudez, 974 F.2d 12, 13 (2d Cir.1992). See infra note 10. . At first glance, Warren v. United States Parole Comm’n, 659 F.2d 183 (D.C.Cir.1981), cert. denied, 455 U.S. 950 , 102 S.Ct. 1454 , 71 L.Ed.2d 665 (1982), appears to hold that such restrictions do not violate the ex post facto clause. See 659 F.2d at 194-95 . However, the holding in Warren rests on the conclusion that the new restrictions do not increase the defendant's punishment. 659 F.2d at 193-94 . The retroactivity discussion is thus dictum. Furthermore, the opinion concedes that the new restrictions "are an incident of [the defendant's] punishment for his first crime, which was committed before the [restrictions] were promulgated.” 659 F.2d at 195 . With this statement, the opinion appears to acknowledge that the restrictions do operate retroactively. . We also note that in many instances courts have not reached the retrospectivity prong of the ex post facto analysis, finding instead that the new restrictions are not more onerous than the old, see, e.g., Conner v. Estelle, 981 F.2d 1032 (9th Cir.1993); Perveler v. Estelle, 974 F.2d 1132 (9th Cir.1992); Mosley v. Klincar, 947 F.2d 1338 (7th Cir.1991); Resnick v. United States Parole Comm’n, 835 F.2d 1297 (10th Cir.1987); Yamamoto v. United States Parole Comm’n, 794 F.2d 1295 (8th Cir.1986), or that the restrictions are not implemented by "laws,'' see, e.g., Kelly v. Southerland, 961 F.2d 1531 (11th Cir.1992); Bailey v. Gardebring, 940 F.2d 1150 (8th Cir.1991), cert. denied sub nom. Bailey v. Noot, - U.S. -, 112 S.Ct. 1516 , 117 L.Ed.2d 652 (1992); Smith v. United States Parole Comm’n, 875 F.2d 1361 (9th Cir.1988); Vargas v. United States Parole Comm'n, 865 F.2d 191 (9th Cir.1988); Sheary v. United States Parole Comm’n, 822 F.2d 556 (5th Cir.1987); Wallace v. Christensen, 802 F.2d 1539 (9th Cir.1986) (en banc); United States ex rel. Forman v. McCall, 776 F.2d 1156 (3d Cir.1985), ce rt. denied sub nom. Forman v. Baer, 476 U.S. 1119 , 106 S.Ct. 1981 ; 90 L.Ed.2d 663 (1986). Here, we are not free to end our analysis before reaching the retroactivity issue. . The Fender court noted that the conduct that serves as the basis for parole revocation — the second offense — may be tried and punished separately as an independent crime and that, for that purpose, any statute adopted prior to the second offense may be applied. It distinguished punishment for the second offense itself from an increase in punishment for the original offense. Id. at 306-07 . See also infra note 10. . As noted above, the date of eligibility for parole is, in these cases, tantamount to a presumptive release date. . We agree that an enhancement statute may escape the strictures of the ex post facto clause if it imposes a “stiffened penalty for the latest crime, considered to be an aggravated offense” because the defendant has committed prior crimes. Gryger v. Burke, 334 U.S. 728, 732 , 68 S.Ct. 1256, 1258 , 92 L.Ed. 1683 (1948) (emphasis added). See United States v. Ahumada-Avalos, 875 F.2d 681 (9th Cir.1989) (holding that an enhancement based on a defendant's criminal record is constitutionally permissible), cert. denied, 493 U.S. 837 , 110 S.Ct. 118 , 107 L.Ed.2d 79 (1989); United States v. Patterson, 820 F.2d 1524 (9th Cir.1987) (same). We note that, in Chatman v. Marquez, 754 F.2d 1531, 1535-36 (9th Cir.1985), cert. denied, 474 U.S. 841 , 106 S.Ct. 124 , 88 L.Ed.2d 101 (1985), we found permissible the California parole board’s enhancement of the term of imprisonment a defendant would be required to serve before being released on parole. Our reasoning in that case, however, did not concern whether the enhancement related to an earlier or a later offense. Rather, our conclusion rested on the ground that the effect of the new parole statute on the petitioner, considered overall, was not more onerous than the old. . The sentence imposed at the time of revocation may not, of course, be more disadvantageous to the defendant than it could have been had it been imposed on the date of the crime. . We note that one case, at first glance, may appear to contain contrary language. United States v. Bermudez, 974 F.2d at 14. In Bermudez , the court vacated the defendant's sentence on an unrelated ground and then suggested in dictum that a violation of supervised release is an offense separate from the underlying crime. However, the court commented upon that question for an entirely different purpose and in an entirely different context. The Bermudez court concluded that, in assessing the severily of the punishment to be imposed for revocation of supervised release, a court should look to the conduct that constituted the violation, rather than to the offense for which the defendant was originally convicted. In that connection, Bermudez pointed out that the offense leading to revocation is an entirely different offense from the underlying crime. We agree with the Bermudez court. The principle it espouses has always obtained in parole cases and is completely harmonious with our holding in this case. In parole revocation, a sentencing court looks to the violation conduct in determining the sentence to be imposed as a result of the violation. As the cases discussed in the text, supra, make clear, however, the court may not exceed the range of punishment that could have been imposed at the time of the underlying offense. Bermudez does not state that a more onerous penalty may be imposed upon revocation than that which could have been imposed at the time of the original offense. Nor does it so much as hint that a court may impose a more severe sentence for revocation of supervised release than for revocation of parole. Bermudez did not involve an ex post facto issue. To the contrary-the appellant argued, and the court agreed, that she was entitled to the lighter sentence applicable under the guidelines rather than being punished by the more severe sentence applicable under the pre-guidelines law according to which she had originally been sentenced. See supra note 4 (explaining that some offenses were punishable by a term of supervised release rather than parole even before the effective date of the Sentencing Reform Act of 1984). The ex post facto clause is in no way implicated when a defendant's sentence is less severe under the new law than under the old. See supra Part II. A. Bermudez is not an ex post facto case. Nothing in Bermudez suggests any view of whether revocation of supervised release is an offense separate from the underlying offense for ex post facto purposes. Nor could the court have suggested any view, given the context in which the issue arose. In short, Bennudez does not purport to speak with respect to any ex post facto issue. Nor do any of the cases it cites. We consider our decision perfectly consistent with Bermudez and those cases. . Instead, the Commission decided that "at revocation the court should sanction primarily the defendant’s breach of trust.” Id. . United States v. Schram, 9 F.3d 741 (9th Cir.1993), is not to the contrary. There, we considered a policy statement, United States Sentencing Commission, Guidelines Manual, § 7B1.3(f), p.s. ("§ 7B1.3(f)”), that states that "any term of imprisonment imposed upon the revocation of ... supervised release shall be ordered to be served consecutively to any sentence that the defendant is serving, whether or not the sentence of imprisonment being served resulted from the conduct that is the basis of ... supervised release.” Id. We concluded that § 7B1.3(f) may be applied to a defendant who committed his underlying crime before the policy statement was adopted and who committed the substantive offense that led to the revocation of his supervised release after the policy statement's effective date. Our holding rested on the acknowledgment that the policy statement implicates the relationship between the penalty for violation of supervised release and the penalty for a new or different offense committed while the defendant is on supervised release. We upheld a rule that a prison term for the violation of supervised release must be served after the completion of any term being served for the new offense. By contrast, 18 U.S.C. § 3583(g), the amendment to the supervised release statute at issue here, implicates the relationship between the penalty for violation of supervised release and the penalty for the underlying crime. Our holding-that, under 18 U.S.C. § 3583 (g), the punishment for a violation of supervised release is a component of the punishment for the underlying crime and cannot be increased retroactively-is perfectly consistent with Schram.