AMENDED OPINION

BOGGS, Circuit Judge.

Defendant Casaneyl Valentine appeals his sentence for committing seven bank robberies on the basis that the district court erroneously departed upward by two levels in determining his offense level under the Sentencing Guidelines. The district court justified the departure on the basis that Valentine robbed seven banks, but, under Section(s) 3D1.4 of the Guidelines, which accounts for multiple groups of offenses, Valentine would only be punished for five of the offenses. Because the Guidelines only allow such departures for "significantly more than five units," and we are convinced by the language and structure of this provision that seven units is not within the range envisioned by the Guidelines as being "significantly more than five," we reverse the decision of the district court, and remand the case for resentencing consistent with this opinion.

I.

Defendant Valentine pled guilty to robbing seven different banks in Memphis, Tennessee, on seven different days between September 17, 1993 and October 12, 1993. Valentine's presentence report established a Guideline range of 130-162 months in prison, based on an offense level of 28 and a criminal history category of V. The presentence report arrived at this range by grouping the seven offenses into seven different groups, taking the offense level of the group with the highest level (26, in this case), and adding five levels pursuant to Section(s) 3D1.4 because Valentine's seven groups of offenses earned him seven "units," bringing him within the category of offenders with more than five units (Section(s) 3D1.4 contains a chart, discussed more fully below, which adds five more levels to an offender's offense level if he or she has "more than five units" in grouped offenses). Valentine qualified for a three-level reduction for acceptance of responsibility, resulting in a final adjusted offense level of 28.

At sentencing, however, the district court sought to correct what it perceived as the injustice of Valentine's last two robberies going "unpunished" pursuant to Section(s) 3D1.4's enhancement table that ends by adding only five levels to the offense level of offenders with more than five units. Accordingly, the district court departed upward by two levels, and sentenced Valentine to 180 months of imprisonment. The district court explained its decision as follows:

As I indicated earlier, my thinking is that the 3D1.4 grouping guideline does not take into account more than five offenses. The defendant has pled guilty to seven offenses, and I think payment is warranted for seven offenses.

* * *

The Court believes that under 3D1.4, the Court should increase the sentence by two additional units which takes into account that there are two felonies. There have been two felony bank robberies that have not been taken into account.

Counsel for Valentine objected to this departure to no avail at the sentencing hearing, and Valentine now appeals the decision to us.

II.

We review decisions to depart from the sentencing guidelines for abuse of discretion. Koon v. United States, 116 S.Ct. 2035, 2046-48 (1996); United States v. Barajas-Nunez, 91 F.3d 826, 831 (6th Cir. 1996). ("The Koon court's abuse of discretion standard replaces the three-part standard of review" previously used by this court). However, an error of law in the interpretation or application of the guidelines constitutes an abuse of discretion. Koon, 116 S.Ct. at 2047-48; Barajas-Nunez, 91 F.3d at 831. Our review compels us to try to reconcile the departure with the language and structure of the Guidelines in much the same manner as we would interpret a statute. Accordingly, since this is a somewhat unusual case for departure wherein the Guidelines specifically note the possible basis for the departure, we start with the plain language of the Guidelines.

The plain language, however, resists a clear answer because it includes the inherently subjective term "significantly more." Specifically, the Guidelines Commentary states:

Inasmuch as the maximum increase provided in the guideline is 5 levels, departure would be warranted in the unusual case where the additional offenses resulted in a total of significantly more than 5 Units.

In unusual circumstances, the approach adopted in this section could produce adjustments for the additional counts that are inadequate or excessive. . . . . Situations in which there will be inadequate scope for ensuring appropriate additional punishment for the additional crimes are likely to be unusual and can be handled by departure from the guidelines. . . . . An alternative method for ensuring more precise adjustments would have been to determine the appropriate offense level adjustment through a more complicated mathematical formula; that approach was not adopted because of its complexity.

U.S.S.G. Section(s) 3D1.4, comment. (backg'd). Thus, the only directive to be gained from the "significantly more" language is that "just plain" more than five is not enough to justify departure — how much more is "significantly more" is unclear.

A careful examination of the chart gives us some evidence for interpreting what could be "significantly more than five." The lowest number of units that could be "more than five" is 5+, but that number is already covered as the very first portion of the category that requires only a five-level increase. The chart is not specific after this point, except for the general sense, inherent in the word "significant," that the magnitude of the difference be more than the bare minimum.

The structure of this Guidelines provision, however, lends considerable support to our holding that, in this context, seven is not "significantly more than five." This provision provides for a progression from "groups of offenses" to "units" to "levels," the basic currency of the guideline scheme. Different offense groups (which can include one or several offenses) are assigned "units," ranging from 0 to + to 1. When we examine the table that is set out in the guidelines, we can see a general pattern that increasingly large numbers of units are required to justify each additional increase in offense level. That is, Section(s) 3D1.4 does not simply add one level for each unit assigned to an offender. Rather, it adds levels as follows:

Thus, even a simple continuation of the pattern would require an increase of a minimum of 2 units to reach the bottom of the range that would justify an increase of six levels. (More likely, there should be an increase of 3 or 4 units, depending on whether the progression of increases is linear or geometric.)

Consistent with this analysis, the hypothetical range that would justify adding six levels would only begin at 7+, and thus Valentine, with his 7 units, would not fall within that range.

An alternative way of looking at the chart is to examine the width of each range; that is, the number of additional units that can be added without an increase in level. Thus, the lowest range, which justifies an increase of one level, has a width of 0: it applies only to exactly 1+ units. The same applies to the range that justifies an increase of two levels. If we look at the entire chart, the widths of the ranges increase as follows: 0, 0, +, 1+. Thus, when we look at the width of the range for the hypothetical increase of five levels, that range should be a minimum of 1+ units, and more likely 2+ units or more, again, depending on whether the rate of increase is linear or geometric. Applying even this hypothetical minimum width of the range that justifies five levels would set that range as being from 5 + to 7 units, and more likely even wider.

Thus, by either means of computation, 7 units would fall only within the range for the adding of five levels. An amount of units that falls comfortably within the range of a simple command to "add five levels" cannot be considered to be "significantly more" than would justify that offense level.

The point we are illustrating here, that Section(s) 3D1.4 attaches decreasing marginal punishment to additional offenses, has been noted by other courts. Moreover, it is explicit or implicit in other schemes found within the Guidelines, and has a strong policy basis. In United States v. Pearson, 911 F.2d 186 (9th Cir. 1990), the Ninth Circuit decided that a district court did not abuse its discretion by departing upward where a defendant accumulated 8 units pursuant to Section(s) 3D1.4. The Pearson court, however, noted the Guidelines' goal of producing declining marginal punishments, and held that the district court's departure to 120 months of imprisonment from the range of 51-63 months was unreasonable in its magnitude. Id. at 190 n. 4. Similarly, the First Circuit took note of the "Sentencing Commission's model of `declining marginal punishments'," yet concluded that an upward departure from a range of 57 to 71 months to 120 months was justified for an offender who pled guilty to 14 counts of bank robbery and one count of attempted bank robbery. United States v. Chase, 894 F.2d 488, 492 (1st Cir. 1990). The Third Circuit decided that a departure was appropriate to account for four victims (equated there with four groups of offenses and, therefore, "units" pursuant to 3D1.4) of a child pornographer not otherwise resulting in additional punishment because of the maximum five-level increase. United States v. MacLeod, 80 F.3d 860, 866 (3rd Cir. 1996). The MacLeod court, however, found that the extent of the departure — four levels for four additional offenses — was unreasonable because it was "at odds with th[e] principle of declining marginal punishment." Id. at 868. The fact that this principle of declining marginal punishment has been clearly enunciated in the introduction to Chapter 3 of the Guidelines and is implicit in the structure of Section(s) 3D1.4, among other provisions, is sufficient to compel courts to give it effect in their sentencing decisions.

While useful in that they support our reliance on the Guidelines' emphasis on declining marginal punishment, the MacLeod decision that a putative 10 units permitted departure and the Pearson decision that 8 units permitted departure are not comparable to this case because their respective courts considered the nature of the underlying offenses as justification for the departures. The MacLeod court considered the four uncounted victims (4 additional units, effectively) significant "[g]iven the potential psychological harm to the young victims of this type of offense. . . ." MacLeod, 80 F.3d at 866. The Pearson court, in affirming a departure based on eight units, noted that the bank robberies at issue "involved the threat of the use of a gun, and occurred in bank areas to which the public had access, enhancing the risk of harm to others." Pearson, 911 F.2d at 189. The Pearson court concluded that the departure was permissible partly because "considering the particular crimes involved and the circumstances of their commission, it properly can be said that Pearson's offenses were unusual." Ibid.

We differ with the other circuits that allow the decision of whether an offender deserves an upward departure pursuant to Section(s) 3D1.4 due to the presence of significantly more than five units to be confused with the general social harm caused by the underlying offenses. The Guidelines established an elaborate system to weigh all, or virtually all, of the facets of an offender's criminal activities. The base offense level assigned to a particular offense generally accounts for the seriousness of the offense, while the sections for specific offense characteristics and the various sections on adjustments for offender and victim characteristics account for these other variables. Section 3D1.4, on the other hand, is meant to account solely for the number of different offenses or groups of offenses that an offender committed. Departure from the chart in this section should thus be based solely on the number of units assigned to an offender, not the underlying nature of the units.

To approach this chart otherwise and interpret its concept of "significantly more than five" to involve some subjective weighing of the social significance of the underlying offenses usurps the role assigned to the Sentencing Commission in setting base offense levels, and turns the section into a catch-all provision justifying departure whenever a court simply believes an offender with more than five units deserves additional punishment. The whole point of the Guidelines is to reduce or remove this type of discretion from the sentencing process and assign certain numerical values to certain facets of an offender's criminal activities. To confound the facet of the Guidelines dealing with the magnitude of criminal activity with other facets of the Guidelines, such as the subjective social harm caused by the particular type of offenses involved, reduces the precision and uniformity of sentences. The type of concerns articulated by the MacLeod and Pearson courts are clearly important, and might justify departure under other, more general provisions of the Guidelines, such as Section(s) 5K2.0. These concerns, however, should not be weighed in the decision of whether an offender has accumulated "significantly more than five units."

III.

Because 7 units is relatively close to the edge of the chart used in 3D1.4 and almost certainly would not have even exceeded the probable next cut-off point had the chart been expanded to include a category for a six-level increase, we conclude, as a matter of law, that 7 units are not "significantly more than 5," so as to permit departure in this context. In so concluding, we note that our mode of analysis was essentially to do that which the Commission itself avoided as too complicated — reconstruct a mathematical formula to handle multiple groups of offenses. Nonetheless, we see this method as the most principled means of answering what the drafters of the Guidelines meant by "significantly more than five." While we cannot point to a particular number of units as the cut-off after which a district court is justified in departing, we can conclude that this point should lie above a line reached by simply extending the general direction of the relationship established in the Guidelines by one more level.

We thus VACATE defendant Valentine's sentence, and REMAND his case to the district court for resentencing consistent with this opinion.