Out of a $10.4 million settlement fund, the district court awarded plaintiffs’ counsel $950,000 in fees and $121,277 in expenses, the total of which — $1,071,277—is about 10.3 percent of the fund.1 On the attorneys’ motion for reconsideration, the court increased the fees to $1,050,000, plus the $121,277 in expenses. Counsel had requested $2,997,000 in fees and $131,900 in expenses. In making the award, the district court relied on the widely accepted lodestar-and-multiplier method and refused to adopt the percentage-of-the-fund approach. The court also declined to award a multiplier. On appeal the attorneys suggest adoption of the percentage-of-the-fund method as a rule of law in common fund cases. They also argue that the district court’s methodology for computing the lodestar figure was improper, that the district court used an improper average for the attorneys’ claimed hourly rates, that class counsel is entitled to a multiplier and that the court improperly denied expenses for computer-assisted legal research.

I.

The underlying litigation was a class action that consolidated several individual suits and a derivative suit against Ly-phomed for alleged violations of section 10(b) of the Securities and Exchange Act of 1934, 15 U.S.C. § 78j(b) (1988). The complaint alleged fraudulent disclosures relating to product development and quality control. In 1987 the Food and Drug Administration was investigating alleged deficiencies in the firm’s quality control measures and had alerted the company to possible violations of the Food, Drug and Cosmetic Act. This case was filed in January 1988. Discovery ensued, substantial settlement negotiations began in November 1988, a memorandum of understanding was reached in April 1989 and the parties entered a settlement stipulation in June 1989. On December 21, 1989, the district court approved the $10.4 million settlement.

Plaintiffs’ counsel then sought fees and expenses from the common fund. In their December 18, 1989, request, counsel asked for $2,997,000 in fees, or just under thirty percent of the total. They also calculated a lodestar of $1,376,423 and suggested a multiplier of 2.177, yielding the same total. The attorneys claimed that the amount was fair and reasonable and that it did not exceed the thirty percent cap promised to class members. Counsel also requested $131,899.93 in expenses. Documentation of the fee request included summaries of expended time, listings of current hourly rates, work descriptions, biographies, *972schedules of expenses and depositions from other attorneys attesting to the rates. The request covered sixty-one attorneys — thirty-one partners, twenty-eight associates and two student interns — representing four Chicago, four New York and two Philadelphia law firms.

The district court’s decision of February 27, 1990, 734 F.Supp. 294, begins with four assumptions. The first is that the substantive reforms that took place at Lyphomed were the result of the FDA enforcement action and not the class action. Second, the court believed that the settlement was not the “extraordinary damage recovery” claimed by counsel. Third, the court considered it a strike against counsel that the settlement had been structured in a way that might hurt class members holding Ly-phomed stock; whether true or not, the court thought it counsel’s duty to mention this aspect of the settlement. Finally, the court observed that the case “did not involve protracted litigation,” although discovery was somewhat vigorously resisted by the defense.

Reviewing the time and rate submissions, the court clearly suspected overbill-ing. The order observes that “a large number of attorneys [are] billing a substantial amount of time and it appears that such circumstances resulted in otherwise unnecessary duplication of effort.” Order at 4. Further, the court requested that counsel provide more detailed billing data for three sample periods of the litigation: preparation of the class certification brief; final settlement approval; and court appearances for twenty-six status and motion calls. The court remained somewhat dissatisfied with the submissions but calculated needed subtotals and totals on its own.

To determine the award, the court first calculated a lodestar. Examining the first of the three sample periods, the judge concluded that class certification preparation showed that “much of counsel’s bill consists of unnecessary duplication of work brought about by having an excessive number of attorneys involved in the case.” Order at 7. The judge reduced the requested $58,591 to $30,000. The court then examined the submissions relating to the memorandum in support of final settlement and decided that the $40,441 billed was overstated; he granted $20,000 as “more than adequate compensation for the memorandum.” Id. Finally, the court detected ov-erbilling in the $24,406 requested for ninety-nine hours of court appearance involving twenty-six status and motion calls. Included in the request were court appearances on dates for which the docket showed no court activity.2 After eliminating the hours of a New York partner the judge thought unwarranted and reducing the rates of a New York associate and two Chicago partners, the court allowed $14,897 of the originally requested $24,406 for appearances. Id. at 8. Based on the 47.4 percent reduction (to $64,897) in the three samplings’ total lodestar request of $123,-438, the court permitted 52.6 percent of the total lodestar requested for the case, netting $783,491.

The judge then took another tack. He assumed that 3,500 hours represented a reasonable number of associate hours for the case, which is close to the requested associate hours of 3,472. He also assumed — on what basis, the opinion does not say — that partners should have spent thirty percent as much time as associates, and that paralegals should have spent fifty percent as much. (This 3:10:5 ratio we will call the “staffing ratio.”) Assuming rates of $225 for partners, $175 for associates and $60 for paralegals as “more than adequate,” the court calculated a lodestar of $236,250 for partners, $612,000 for associates and $105,000 for paralegals (total = $953,750). The court believed this a bit high and rounded the figure to $950,000, which the court accepted as the lodestar for the case.

The court then assessed the arguments for a muliplier. Counsel contended that a multiplier of more than two was appropri*973ate to attract counsel to contingent, complex litigation of this type. The court concluded that “[s]uch an argument is ludicrous as applied to this case” because plenty of attorneys had found the case worth their time. Id. at 10. Also believing that the quality and complexity of the work had been compensated by the high hourly rates and number of hours, the court denied any multiplier.

The court then considered the attorneys’ suggestion to use the percentage-of-the-fund method for calculating the fees. The attorneys argued that the percentage method was simpler and less time-consuming and avoided the lodestar’s incentive to bill excessive hours. But the court believed the lodestar-and-multiplier approach manageable to apply and found unpersuasive the fact that the lodestar method generally results in awards of twenty to thirty percent. The court also found undesirable incentive features in the percentage method: “A percentage award provides incentive to the attorneys to settle relatively cheaply early in the litigation since it can provide them with a larger hourly fee and then free them up to pursue more cases that they can also settle early in the litigation for less than the best value for the class.” Id. at 13 (footnote omitted). The court concluded that in the absence of controlling precedent, it would continue to apply the lodestar-and-multiplier method. As an afterthought, the court remarked that even if required to use the percentage method, a percentage lower than the traditional 20 percent floor would be appropriate in view of the relatively few risks in this case.

The court also granted counsel’s request for $131,900 in expenses, except for $10,-623 in computer-assisted research fees, resulting in a total of $121,277. The court believed the computer expense was adequately reflected in the attorney’s fees.

On the attorneys’ motion for reconsideration, the court increased the hourly rates used in the staffing ratio-based lodestar to the averages of those requested ($247 for partners, $190 for associates and $71 for paralegals), which increased the lodestar by $100,000. The court did not alter its judgment regarding a multiplier and still excluded the computer-assisted research expense. The final award was therefore $1,050,000 in fees and $121,277 in expenses, plus interest.

II.

A. Standard of Review

Our standard of review in fees cases is a well-established abuse of discretion standard. Dutchak v. Central States, S.E. & S.W. Areas Pension Fund, 932 F.2d 591, 596 (7th Cir. 1991); Skelton v. General Motors Corp., 860 F.2d 250, 257 (7th Cir. 1988), cert. denied, 493 U.S. 810, 110 S.Ct. 53, 107 L.Ed.2d 22 (1989); Ustrak v. Fairman, 851 F.2d 983, 987 (7th Cir. 1988). District courts are far better suited than appellate courts to assess a reasonable fee in light of the case’s history. Abuse of discretion occurs when the court reaches an erroneous conclusion of law, fails to explain a reduction or reaches a conclusion that no evidence in the record supports as rational. We will also review de novo the district court’s methodology to determine whether it reflects procedure approved for calculating awards.

B. Percentage-of-the-Fund Method

The lodestar method, the most widely cited description of which is found in Lindy Bros. Builders, Inc. v. American Radiator & Standard Sanitary Corp., 487 F.2d 161, 167-69 (3d Cir. 1973) {Lindy I), appeal following remand, 540 F.2d 102 (3d Cir. 1976) {Lindy II), is regularly used in statutory fee shifting cases, see Hensley v. Eckerhart, 461 U.S. 424, 432-34, 103 S.Ct. 1933, 1938-40, 76 L.Ed.2d 40 (1983), and is frequently employed in common fund cases, see In re Folding Carton Antitrust Litigation, 84 F.R.D. 245 (N.D.Ill. 1979). In a common fund case the judge exercises equitable powers to compensate attorneys from the recovery won for plaintiffs. Alyeska Pipeline Serv. Co. v. Wilderness Soc’y, 421 U.S. 240, 257-58, 95 S.Ct. 1612, 1621-22, 44 L.Ed.2d 141 (1975).

The district court declined to use the percentage method in making this award. *974Class counsel argue that the percentage method offers advantages that recommend it over the lodestar method in common fund cases. They argue that the lodestar method is inefficient and burdensome, citing cases where the method has been criticized. See In re Activision Sec. Litigation, 723 F.Supp. 1373 (N.D.Cal. 1989) (thirty percent recovery permitted); In re Union Carbide Corp. Consumer Prod. Business Sec. Litigation, 724 F.Supp. 160, 170 (S.D.N.Y. 1989) (employing lodestar approach but anticipating return to contingency-type awards) [; see also Court Awarded Attorney Fees, Report of the Third Circuit Task Force, 108 F.R.D. 237, 254-59 (1985)]. At least one circuit has approved use of the percentage method in a common fund situation. Paul, Johnson, Alston & Hunt v. Graulty, 886 F.2d 268 (9th Cir. 1989). But cf. Florida v. Dunne, 915 F.2d 542 (9th Cir. 1990) (percentage or lodestar method approved). Second, they contend that the percentage method is an efficient and reasonable way to compute fees, since it requires simply discerning the appropriate percentage from similar cases and multiplying. Third, they argue that even if the lodestar method is required for fees awarded under a statutory shifting provision, it need not be used in common fund situations.

Although the lodestar approach certainly has problems, we think it premature to banish it now. First, the lodestar process initially responded to concerns that a percentage approach resulted in over-compensation for attorneys. Requiring an explicit accounting of hours and rates provides greater accountability. Second, the separation of multiplier and lodestar provides — in theory at least — a means of accounting both for the hours and rate reasonable in the type of case and for the risk an attorney assumes in undertaking a case.

In addition, the lodestar encourages attorneys to assess the marginal value of continuing to work on the case. Assured of recovering a “reasonable fee” if victorious, an attorney will not find it worthwhile to spend time that does not add to the value of the suit, since there is significant risk that the court will in the end disapprove such time. A frequently cited advantage to the percentage method — its alignment of class attorneys’ interests with those of the client — may, as the district court noted, encourage holding out or early settlement on the basis of the expected value of the recovery. These incentive effects are, of course, subject to many other pressures in the actual conduct of litigation, but we think that they indicate at least that the door should not be closed on lodestar awards.

We therefore affirm the traditional method for calculating fees. The district court should determine the reasonable number of hours expended by the attorneys. “Reasonable” implies that the judge may reduce the hours if the number requested appears excessive compared to other cases or in light of the work. Next, the judge should determine the reasonable hourly rate allowable for each attorney or class of attorneys involved in the case. These rates should be within narrow bands of rates charged by attorneys of similar experience, competence and specialty. Multiplying the hours reasonably expended by the reasonable hourly rates produces the lodestar.

Then comes the issue of multipliers. Judges describe the need for multipliers in a variety of ways, but these explanations have a common theme — compensating in a manner that provides adequate incentive for the attorney to bring this type of case. This means that the judge should view the multiplier from an ex ante perspective; that is, what size risk the attorney assumed at the outset by taking this type of case. Skelton, 860 F.2d at 258 (multiplier computation involves “retroactive calculation of the probability of success as measured at the beginning of litigation”). This requires comparing the attorney’s prospect for success with other cases of similar dimensions.

After establishing the lodestar and multiplier, the judge may find it useful to compare the percentage of the fund to contingent arrangements negotiated in other cases of the same type. Because the target is an award that reflects a worthwhile ex ante arrangement for counsel, this final step is a reasonable, though not essential, “check” for the judge.

*975We do not think the judge abused his discretion by rejecting the percentage approach. While we do not rule out use of the percentage method when the district court finds it appropriate, see Gekas v. Attorney Registration & Disciplinary Comm., 793 F.2d 846, 851-52 n. 2 (7th Cir. 1986), neither do we require its use in all common fund cases. The decision of whether to use a percentage method remains in the discretion of the district court. Kirchoff v. Flynn, 786 F.2d 320, 329 n. 1 (7th Cir. 1986).

C. Challenges to the District Court’s Award

1.The District Court’s Methodology

The attorneys challenge the district court’s use of the staffing ratio. They claim that it bears no relation to the actual hours expended by the three classes of attorneys in the case and also reflects a bias toward (generally large) law firms, which operate with lower partner/associate staffing ratios.

We think that the staffing ratio was an inappropriate procedure for determining the size of the lodestar. The district court’s further assumption of 3,500 associate hours is in line with the requested number (3,472), but this seems inconsistent with the judge’s earlier conclusion from the samplings that time had been overstated. We find no support in the record for the proposition that the staffing ratio mirrors patterns in the case or litigation practice generally. The starting point for the district court’s analysis must be the reasonable number of hours the attorneys claim actually to have spent. If the hours submitted for a particular category of attorneys or a segment of work appear high, the court should adjust them accordingly. Because the district court provided inadequate explanation of the basis for the staffing ratio, it cannot support the award of fees.

That said, we find no error in the district court’s use of a “sampling” procedure to estimate the number of hours appropriate for compensation. The court selected three significant segments of the litigation that it felt reflected typical attorney work in the case. It then made findings regarding the attorney billing practices and thought this likely represented billing practices for the entire litigation. The judge’s chosen procedure is a bit perplexing. The admirable precision with which the sampling was conducted appears to have been abandoned for the more amorphous staffing ratio calculations. If the lodestar fails to produce a just fee award, the court may either adjust the multiplier based on the risk involved or set the award on a percentage basis.

2. Hourly Rates

The attorneys also dispute the “average” rates used by the judge in computing the lodestar. Specifically, they claim that by using the average rates for partners, associates and paralegals in the case, the judge committed legal error by not taking into account regional and experience differences that affect the rates of individual attorneys.

We do not find an abuse of discretion in the judge’s calculation. The “average” rates that he accepts in the award granted on the motion for reconsideration are not hypothetical national averages, as was involved in In re Fine Paper Antitrust Litigation, 751 F.2d 562, 590-91 (3d Cir. 1984). They represent the average rates requested by the classes of attorneys in the case— for example, total lodestar for partners divided by hours requested by all partners. The averaging reduces the overall compensation a bit,3 but counsel appear more concerned that the award does not break down the award by attorney. While this may have been our preferred method for toting up the fees, it was not an abuse of discretion for the district court to leave this job to the attorneys themselves.

3. The Multiplier

Counsel contend that the district court erred in not awarding a risk *976multiplier. A multiplier is, within the court's discretion, appropriate when counsel assume a risk of non-payment in taking a suit. We agree that the counsel is probably entitled to a contingent multiplier, as there was some probability that the suit would not result in success for the plaintiffs. The judge’s conclusion that the case was not particularly risky does not indicate whether this is an ex ante or ex post determination. Moreover, the judge’s conclusion that “the element of complexity is adequately compensated by.the fact that increased complexity results in increased hours which produce a higher lodestar,” Order at 10-11, misses the point that the multiplier is designed to reflect the fact that, no matter how many hours were invested, there was, at the outset, the possibility of no recovery. See Skelton, 860 F.2d at 258 (discussing rationale for risk multiplier and method of assessing). We therefore must remand for reconsideration of the multiplier issue and ask the judge to assess the multiplier with regard to the probability of success in this type of litigation at the outset of the case. Multipliers anywhere between one and four, In re Cenco, Inc. Sec. Litigation, 519 F.Supp. 322, 325 (N.D.Ill. 1981) (four, which is quite rare), have been approved. The level selected should, to the extent possible, be without regard to most developments during discovery and litigation because it is designed to reflect the riskiness of the case at the outset. Lindy Bros., 540 F.2d at 117. Subsequent developments affecting the case should be reflected in the hours or rates thought reasonable.

D. Computer-Assisted Research

The attorneys also claim that the court erred in excluding the roughly $10,-000 expended in computer-assisted research. The court felt that the expense “is part of the amount allowed for attorney’s fees.” We must reverse the district court here. Computer-assisted research fees — so long as reasonably incurred — in theory reduce the number of attorney hours otherwise needed for (presumably) more time-consuming manual research and are therefore compensable. It is, of course, within the judge’s discretion to find the expenses unreasonably incurred.

III.

The district court fee award is Vacated and Remanded for recalculation consistent with the discussion above. The district court's disallowance of the computer-assisted research expense is Reversed.