This case presents the question whether — and if so, how — a district court may permissibly reduce an attorney’s fees award under federal fee-shifting statutes to reflect poor quality of representation.
FACTUAL And PROCEDURAL BACKGROUND
Plaintiff Maria Van Gerwen sued defendants under ERISA § 502 for improperly denying her past and future benefits under an ERISA long term disability plan. Following discovery, the parties filed cross-motions for summary judgment and the district court granted Van Gerwen’s motion, ruling that Van Gerwen was entitled to benefits under the plan as well as $40,-412 in damages.
Van Gerwen moved for attorney’s fees for 127.25 hours at a rate of $300 an hour — a total of $38,175.1 The court awarded only $14,212.50 in fees. Relying on defendants’ evidence that the customary hourly rate for an ERISA plaintiffs attorney is “no more than $200 per hour” and Van Gerwen’s evidence that the range of hourly rates for an ERISA plaintiffs attorney is $175-350, the district court concluded $200 was a reasonable hourly rate. The court then awarded fees for only 94.75 hours, because it found that only the administrative record was relevant evidence and that Van Gerwen’s attorney unnecessarily spent hours on discovery unrelated to that record. Finally, the district court reduced the total amount of attorney’s fees by a multiplier of .75 based on the poor quality of the attorney’s work. The court reasoned that the “Statement of Genuine Issues” opposing defendants’ motion for summary judgment submitted by Van Gerwen’s attorney contained no citations to evidentiary material as required by Federal Rule of Civil Procedure 56(e) and Local Rule 7.14.2. Moreover, the court noted the attorney’s filings in support of summary judgment did not cite to the administrative record or any other evidentiary material. The court concluded the attorney’s performance “imperiled his client’s likelihood of prevailing in this case” and therefore reduced the award of attorney’s fees accordingly.
Van Gerwen contends the district court erred in relying on quality of representation in applying the .75 multiplier. She also argues the court erred in deducting hours from her request for time spent on discovery unrelated to the administrative record and for ensuring the integrity of the record, and the court improperly considered the contingency nature of her fee agreement with the attorney in determining the reasonable hourly rate for the attorney’s services.' We vacate the district court’s attorney’s fee determination because, on the record before us, it appears the district court may have erroneously relied on quality of representation both in calculating the lodestar amount and in applying a downward multiplier.
*1045STANDARD Of REVIEW
Under ERISA § 502(g), a district court has discretion to award reasonable attorney’s fees. See 29 U.S.C. § 1132(g)(1) (1994). We review for abuse of discretion a district court’s decision 'to award or deny attorney’s fees in an ERISA action, see Friedrich v. Intel Corp., 181 F.3d 1105, 1113 (9th Cir.1999); McElwaine v. U.S. West, Inc., 176 F.3d 1167, 1171 (9th Cir.1999), as well as a district court’s determination of the amount of reasonable attorney’s fees. See D’Emanuele v. Montgomery Ward & Co., 904 F.2d 1379, 1384 (9th Cir.1990); Sapper v. Lenco Blade, Inc., 704 F.2d 1069, 1073 (9th Cir.1983).
DISCUSSION
This court has adopted the hybrid lodestar/multiplier approach used by the Supreme Court in Hensley v. Eckerhart, 461 U.S. 424, 103 S.Ct. 1933, 76 L.Ed.2d 40 (1983), as the proper method for determining the amount of attorney’s fees in ERISA actions. See McElwaine, 176 F.3d at 1173; D’Emanuele, 904 F.2d at 1383. The lodestar/multiplier approach has two parts. First, a court determines the “lodestar” amount by multiplying the number of hours reasonably expended on the litigation by a reasonable hourly rate. See D’Emanuele, 904 F.2d at 1383; Hensley, 461 U.S. at 433, 103 S.Ct. 1933. The party seeking an award of fees must submit evidence supporting the hours worked and the rates claimed. See Hensley, 461 U.S. at 433, 103 S.Ct. 1933. A district court should exclude from the lodestar amount hours that are not reasonably expended because they are “excessive, redundant, or otherwise unnecessary.” Id. at 434, 103 S.Ct. 1933. Second, a court may adjust the lodestar upward or downward using a “multiplier” based on factors not subsumed in the initial calculation of the lodestar.2 See Blum v. Stenson, 465 U.S. 886, 898-901, 104 S.Ct. 1541, 79 L.Ed.2d 891 (1984) (reversing upward multiplier based on factors subsumed in the lodestar determination); Hensley, 461 U.S. at 434 n. 9, 103 S.Ct. 1933 (noting that courts may look at “results obtained” and other factors but should consider that many of these factors are subsumed in the lodestar calculation). The lodestar amount is presumptively the reasonable fee amount, and thus a multiplier may be used to adjust the lodestar amount upward or downward only in “ ‘rare’ and ‘exceptional’ cases, supported by both ‘specific evidence’ on the record and detailed findings by the lower courts” that the lodestar amount is unreasonably low or unreasonably high. See Pennsylvania v. Delaware Valley Citizens’ Council for Clean Air, 478 U.S. 546, 565, 106 S.Ct. 3088, 92 L.Ed.2d 439 (1986) (quoting Blum, 465 U.S. at 898-901, 104 S.Ct. 1541); Blum, 465 U.S. at 897, 104 S.Ct. 1541; D’Emanuele, 904 F.2d at 1384, 1386; Cunningham v. County of Los Angeles, 879 F.2d 481, 487 (9th Cir.1988).
*1046I. The District Court’s Reliance On Quality Of Representation To Reduce The Amount Of Attorney’s Fees By A Multiplier Of .75
Van Gerwen argues the district court abused its discretion by using a .75 multiplier to reduce the amount of attorney’s fees below the lodestar amount, improperly using the multiplier to factor in the quality of her attorney’s representation, a factor already accounted for in calculating the lodestar. On the record as it now stands, we agree.
Quality of representation is generally considered at the lodestar stage in determining what is a reasonable hourly rate. See Blum, 465 U.S. at 899, 104 S.Ct. 1541. To factor quality of representation also into the multiplier risks double counting. In Blum, the Court rejected use of a multiplier to increase the fee award where the district court had already awarded fees at the upper end of the range based on factors that included the high quality of the attorneys’ representation. Id. at 899-90, 104 S.Ct. 1541. Blum recognized that in rare cases the lodestar amount might be unreasonable and considerations such as “the important factor of the results obtained” might lead a district court permissibly to adjust a fee upward or downward. Id. at 897 & n. 14, 104 S.Ct. 1541. Nonetheless, the Court made clear (as have we) that, absent exceptional circumstances, it is impermissible to use the multiplier to adjust the lodestar fee for quality of representation. In the • context of an upward adjustment, as in Blum, such an adjustment is justified only in the rare case where there is specific evidence that the quality of service was superior in light of the hourly rates charged and the success was exceptional. Id. at 899, 104 S.Ct. 1541; see also Delaware Valley, 478 U.S. at 567-68, 106 S.Ct. 3088 (reversing upward multiplier because lower court did not specifically find that quality of representation was not reflected in lodestar and party presented no evidence that result was outstanding).
By the same token, a district court may not rely on quality of representation to apply a downward multiplier unless, at the very least, the district court makes a detailed finding based on specific evidence that the quality of service was inferior in fight of the hourly rate selected in setting the lodestar amount. In Cunningham, we concluded quality of representation cannot be considered at the multiplier stage, “at least in the absence of specific evidence that the attorney’s performance is exceptional or abysmal.” 879 F.2d at 487 (citing Blum, 465 U.S. at 899-900, 104 S.Ct. 1541, and Delaware Valley, 478 U.S. at 566-67, 106 S.Ct. 3088); see also D’Emanuele, 904 F.2d at 1384 (concluding that under Blum, quality of representation generally should not be relied upon to multiply lodestar amount “upward or downward.” (emphasis added)).
In the present case, the district court did not explain whether or to what extent it considered quality of representation when it determined the lodestar amount. We infer from the court’s selection of an hourly rate at the low end of the range submitted by Van Gerwen that the $200 rate reflected the court’s assessment of the attorney’s quality of representation. See Delaware Valley, 478 U.S. at 565, 106 S.Ct. 3088 (concluding that quality of representation is “presumably fully reflected in the lodestar amount”). We affirmatively know from the district court’s findings that it did consider quality of representation at the multiplier stage. In arriving at the .75 multiplier, the district court cited substantial evidence of the attorney’s poor quality of representation, such as his failure to cite to evidentiary material either in support of Van Gerwen’s motion for summary judgment as required by Federal Rule of Civil Procedure 56(e) and Local Rule 7.14.2 or in opposition to the defendants’ motion. See, e.g., Nilsson, Robbins, Dalgarn, Berliner, Carson & Wurst v. Louisiana Hydrolec, 854 F.2d 1538, 1545 (9th Cir.1988). Although we are sympathetic to the district court’s frustration with counsel, the court did not specifically find that the at*1047torney’s performance was so abysmal that the lodestar amount was unreasonable, nor did it explain why the hourly rate it selected did not fully account for the quality of representation. See Delaware Valley, 478 U.S. at 565, 106 S.Ct. 3088; Cunningham, 879 F.2d at 487. Absent the requisite detailed findings, it áppears the district court may have erroneously double counted the quality of performance.
The concern with double counting is to avoid assessing a plaintiff and her attorney “a double penalty” for the poor performance, thereby awarding defendants an unjustified windfall by requiring them to pay less than a reasonable fee amount. Cunningham, 879 F.2d at 488. Moreover, double counting quality of performance would ignore that the main purpose of fee-shifting statutes is not to punish or reward attorneys. Rather, “the aim of such statutes [i]s to enable private parties to obtain legal help in seeking redress for injuries resulting from the actual or threatened violation of specific federal laws.” Delaware Valley, 478 U.S. at 565, 106 S.Ct. 3088.
If this is a rare case where it is not possible to address fully the attorney’s performance by setting the appropriate, reasonable hourly rate used in calculating the lodestar, then the district court must explain why that is so. We therefore remand to the district court to determine, and to explain how, counsel’s service— particularly given the results obtained— relates to the service and results reasonably expected of an attorney working the number of hours at the hourly rate the district court sets in determining the lodestar.3
II. Subtraction Op Hours For Discovery Unrelated To The Record
We reject Van Gerwen’s two other arguments regarding the court’s fee award. First, Van Gerwen argues the district court improperly deducted roughly 27 hours from her attorney’s fees request on the ground that discovery unrelated to the administrative record was unnecessary. Van Gerwen contends most of the 27 hours were spent on discovery aimed at ensuring the integrity of the administrative record and thus were related to the record. As for the remainder, Van Gerwen argues counsel was searching in good faith for alternative arguments to support her claim.
We accord considerable deference to. the district court’s findings regarding whether hours claimed by prevailing counsel are excessive, redundant or otherwise unnecessary. See McGrath v. County of Nevada, 67 F.3d 248, 255 (9th Cir.1995); Davis, 976 F.2d at 1544; see also Hensley, 461 U.S. at 437, 103 S.Ct. 1933. We do so due to “the district court’s superior understanding of the litigation and the desirability of avoiding frequent appellate review of what essentially are 'factual matters.” Hensley, 461 U.S. at 437, 103 S.Ct. 1933. The district court need only provide a “concise but clear explanation of its reasons” for reducing the number of hours included in the fee award. Id.
In an ERISA case, a district court reviewing a plan administrator’s decision to deny benefits may consider evidence outside the administrative record “only when circumstances clearly establish that additional evidence is necessary to conduct an adequate de novo review of the benefit decision.” Mongeluzo v. Baxter Travenol *1048Long Term Disab. Benefit Plan, 46 F.3d 938, 944 (9th Cir.1995). Here, the district court determined that evidence outside the record was neither admissible nor necessary to litigate the case. Van Gerwen has hot' explained why it was necessary to dedicate a substantial amount of time to ensure the integrity of the record or why evidence outside the record was necessary to enable the court to perform adequate de novo review. Indeed, even during discovery, Van Gerwen’s attorney acknowledged it was highly unlikely any evidence outside /of the administrative record would be admissible.
Van Gerwen is correct that at the discovery stage an attorney cannot be absolutely sure whether discovery unrelated to the administrative record will be necessary to litigate a case. However, Van Gerwen’s claim of good faith is belied by her attorney’s acknowledgment that there was little chance discovery unrelated to the record would be admissible. Therefore, the district court did not abuse its discretion in deducting roughly 27 hours from Van Ger-wen’s request for attorneys fees.
III. The District Court’s Statements About The Contingency-Fee Agreement
Van Gerwen also argues the district court improperly relied on the contingency-fee agreement to reduce the amount of the fee award when the court noted counsel was working under:
A “straight contingency basis” [which] presumably entitles [the attorney] to a percentage, as much as one-third, of the present value of the award to the prevailing plaintiff. In this case, plaintiffs counsel seeks not merely one third, but an amount almost equal to the total awarded to the plaintiff (not counting the future benefits to which she may continue to be entitled).
A district court may not rely on a contingency agreement to increase or decrease what it determines to be a reasonable attorney’s fee. See Davis v. City & County of San Francisco, 976 F.2d 1536, 1548-49 (9th Cir.1992), vacated in part on other grounds, 984 F.2d 345 (9th Cir.1993); Quesada v. Thomason, 850 F.2d 537, 543 (9th Cir.1988) (holding that district court abused its discretion in applying downward multiplier to lodestar amount based on contingency agreement); see also City of Burlington v. Dague, 505 U.S. 557, 566-67, 112 S.Ct. 2638, 120 L.Ed.2d 449 (1992) (holding that federal fee-shifting statutes do not allow for upward adjustments to lodestar amount based on contingency agreement). In this case, however, the district court’s comments about the contingency-fee agreement appear to have been merely a summary of the evidence Van Gerwen submitted — triggered by Van Ger-wen’s having raised the issue — and were in any event harmless because the court plainly relied on permissible factors that independently justify its lodestar determination. Specifically, the district court relied on declarations submitted by defendants and by Van Gerwen showing that $200 was a reasonable hourly rate for an ERISA plaintiffs attorney in the community with similar skill and experience. See Davis, 976 F.2d at 1547 (concluding declarations of prevailing market rate in relevant community are sufficient to establish appropriate rate for lodestar purposes). Moreover, as discussed above, the district court relied on the fact that Van Gerwen’s attorney unnecessarily spent hours pursuing discovery of evidence unrelated to the record.
The present case is therefore distinguishable from Davis, where the district court not only mentioned the contingency agreement but specifically listed it as a factor to be considered in setting the lodestar. See id. at 1549. In Davis, we concluded that even though the district court’s decision appeared to rest on other factors, it was “appropriate that on remand the district court consider whether it would arrive at the same lodestar figure ... without taking the factor of contingency into account.” Id. In this case, unlike Davis, the district court did not specify contingency as a factor to consider in de*1049termining the lodestar. Therefore, we conclude the district court did not abuse its discretion, because it relied solely on permissible factors in making the lodestar determination.
CONCLUSION
We hold that the district court abused its discretion by adjusting downward its award of attorney’s fees at the multiplier stage to reflect poor quality of representation without providing the requisite explanation why quality of representation was not accurately reflected in the lodestar amount. We further hold that the district court did not abuse its discretion in refusing to award fees for hours spent on discovery unrelated to the record or in mentioning, without relying on, the contingency-fee agreement. We therefore vacate the district court’s fee award and remand for a redetermination consistent with this opinion.
VACATED and REMANDED.
. Van Gerwen claimed six hours for the motion for attorney’s fees. Apparently the district court did not include those hours in the fee award. Van Gerwen does not challenge this aspect of the fee award.
. Under Hensley, 11 factors are relevant to the determination of the amount of attorney's fees: (1) the time and labor required; (2) the novelty and difficulty of the issues; (3) the skill requisite to perform the legal service properly; (4) the preclusion of employment by the attorney due to acceptance of the case; (5) the customary fee; (6) time limitation^ imposed by the client or the circumstances; (7) the amount involved and the results obtained; (8) the experience, reputation and ability of the attorneys; (9) the 'undesirability” of the case; (10) the nature and length of the professional relationship with the client; and (11) awards in similar cases. 461 U.S. at 430 n. 3, 103 S.Ct. 1933. Under City of Burlington v. Dague, 505 U.S. 557, 112 S.Ct. 2638, 120 L.Ed.2d 449 (1992), the contingency nature of a fee cannot be considered as a factor bearing on the amount of attorney's fees: See Davis v. City & County of San Francisco, 976 F.2d 1536, 1546 n. 4 (9th Cir.1992), vacated in part on other grounds, 984 F.2d 345 (9th Cir.1993); see also id. (suggesting also that Dague casts doubt on the relevance of "undesirability” to the fee calculation). Further, the Supreme Court has held that novelty or complexity of the issues should not be considered at the multiplier stage, and that quality of representation and results obtained are ordinarily subsumed in the lodestar determination and in most cases should not be considered in the multiplier step. See Blum v. Stenson, 465 U.S. 886, 898-901, 104 S.Ct. 1541, 79 L.Ed.2d 891 (1984).
. If the district court did not consider quality of representation fully or at all in its lodestar determination, it presumably could have done so by setting the hourly rate used in the lodestar at a lower point in the range of customary fees for an ERISA plaintiff's attorney in the relevant community. Of course, a district court also has the authority to sanction an attorney directly for various kinds of poor performance, such as violating local rules, which appears to be the kind of deficient performance that concerned the court here. See, e.g., C.D. Cal. Loc. Civ. R. 7.19; Fed. R. Civ. P. 11, 26(g)(3), 37.