We consider here the fairness and reasonableness of a settlement in an environmental action brought under CERCLA.1 Appellants challenge the settlement (involving other parties) because, they claim, the district court did not have sufficient information before it to properly determine whether the decree— which provided for a $45.7 million settlement with the Los Angeles County Sanitation District and approximately 150 local governmental entities — was “reasonable, fair, and consistent with the purposes that CERCLA is intended to serve.” See United States v. Cannons Engineering Corp., 899 F.2d 79, 85 (1st Cir.1990) (quoting H.R.Rep. No. 253, Pt. 3, 99th Cong., 1st Sess. 19 (1985), reprinted in 1986 U.S.Code Cong. & Admin.News 3038, 3042).
Because we find that the district court could not adequately evaluate the fairness and reasonableness of the proposed consent decree without having before it at least an estimate of the projected total natural resource damages at issue in this ease, we vacate the district court’s approval of the consent decree and remand the matter for the court to determine whether — in light of the governments’ best estimate of projected total natural resource damage, as well as the numerous other factors properly considered in the evaluation of a settlement of this type — the proposed consent decree is fair, reasonable and consistent with the objectives of CERCLA.
BACKGROUND
1. The Lawsuit
In a two-count complaint filed in 1990, the United States and the State of California (“the governments”) sued several corporations and the County Sanitation District No. 2 of Los Angeles County (“LACSD”) seeking natural resource damages and response costs under CERCLA.
The governments’ first claim for relief sought damages pursuant to 42 U.S.C. § 9607(f)(1) for destruction of, injury to, or loss of natural resources in the waters off the eoast of Los Angeles. Those waters include the San Pedro Channel and the area surrounding the Palo Verdes Peninsula, the Los Angeles-Long Beach Harbors, and the waters off Santa Catalina Island and the Channel Islands. The governments allege that natural resource damages in these areas were caused by the release of hazardous substances, including DDT and PCBs. The *744release of these hazardous substances dates back to the late 1940s.
The second claim for relief, brought by the United States on behalf of the EPA, sought recovery of response costs for clean-up of the Montrose Chemical manufacturing site in Torrance, California. Only those corporations associated with the Montrose site were defendants under the second claim.
2. The Parties
(a)The Corporate Defendants
The natural resource damages claim identified nine separate corporate defendants charged with releasing DDT and PCBs into the affected areas. These nine defendants fall into three distinct groups: the Montrose Defendants, Westinghouse, and the Potlatch-Simpson Defendants.
(i)The Montrose Defendants
Montrose Chemical Corporation of California (“Montrose”) operated a DDT manufacturing plant in Torrance, California from 1947 to 1982. Montrose allegedly disposed of process waste containing DDT through ocean dumping, discharge into the sewer collection system operated by LACSD, surface water runoff from its plant site into the Los Ange-les/Long Beach Harbors, and aerial dispersal of waste from its manufacturing operations. The governments claim that Montrose released as much as 5,500,000 pounds of DDT into the affected areas. See U.S. v. Montrose Chemical Corp. of California, 793 F.Supp. 237, 240 (C.D.Ca.1992) (Montrose I).2
(ii)Westinghouse
Westinghouse Electric Corporation (‘Westinghouse”) is also identified as a defendant. Westinghouse operated an electrical transmission equipment maintenance and repair plant in Los Angeles from the late 1950s until its recent closing. The governments charged Westinghouse with the release of wastewater containing PCBs into the LACSD and, in turn, into the Pacific Ocean. LACSD reports indicate that Westinghouse discharged as much as 38,000 pounds of PCBs into the LACSD system. Id.
(iii)The Potlatch-Simpson Defendants
Potlatch Corporation, which operated a paper manufacturing plant in Pomona, California from 1952 to 1979, was charged with releasing wastewater containing PCBs into the LACSD system. Simpson Paper Company purchased the Potlatch facility in 1979 and continued Potlatch’s wastewater disposal practices. The governments estimate that the Potlatch-Simpson defendants released as much as 4,500 pounds of PCBs into the affected areas. Through a consent decree similar to the one at issue in this appeal, the Potlatch-Simpson defendants settled with the governments in 1992.
(b)LACSD
The governments also named LACSD as a defendant. LACSD is a publicly-owned sewage system that discharges treated effluent into the Pacific Ocean. The governments allege that LACSD received wastewater from the Montrose, Westinghouse and Pot-latch-Simpson defendants and did not remove all the DDT and PCBs from the waste-water it discharged.
(c)The Third-Party Local Government Entity Defendants
In response to the governments’ lawsuit, the corporate defendants filed third-party claims against 150 local government entities. The local government entities are cities, counties and other governmental bodies that, according to the corporate defendants, discharge various substances, including pollutants, into the affected area. Specifically, the corporate defendants allege that the local governmental entities owned or used sanitation systems and storm runoff systems that *745discharged wastewater into the ocean, or otherwise engaged in activities (such as mosquito abatement) that may have resulted in the discharge of hazardous substances, like DDT, into the environment.
3. The Special Master and Settlement Discussions
The district court appointed Harry V. Pee-tris, a retired Los Angeles Superior Court Judge, to act as Special Master for the lawsuit to supervise all non-dispositive pretrial proceedings and to conduct and supervise settlement negotiations. The Special Master divided the defendants into four groups for purposes of settlement discussions: (1) Montrose, (2) Westinghouse, (3) Potlatch-Simpson, and (4) LACSD and the 150 third-party local governmental entities. Each of the four groups met separately with the governments and the Special Master, and the defendants were prohibited by a confidentiality order from discussing among themselves the substance of their settlement negotiations. See Montrose I, 793 F.Supp. at 241.
At the Special Master’s instruction, the governments created and disclosed to the Special Master alone an overall monetary framework for early settlement of the entire litigation, including specific allocations among the defense groups, based on the governments’ estimates of potential damages and individual liability, and taking into account the costs and risks of litigation.
The negotiations over which the Special Master presided yielded settlements with two defense groups: (1) the Potlatch-Simpson defendants, and (2) LACSD and the local governmental agencies. On May 19, 1992, the district court approved a consent decree, similar to the decree at issue in this appeal, through which Potlatch-Simpson agreed to pay $12 million to the governments in exchange for covenants not to sue and recognition of Potlatch-Simpson’s release from any claims for contribution. See id. The district court’s approval of the Potlatch-Simpson consent decree is not challenged on appeal.
As he had done with the Potlatch-Simpson decree, the Special Master recommended to the district court that it approve the LACSD consent decree. In a twelve-page report, the Special Master concluded, among other things, that the settlement was reasonable given the available information concerning the nature of the activities giving rise to the claimed liability, the settler’s relationship to the damages and response costs alleged, the estimates of restoration and response costs, the litigation risks, and the relative benefits to the parties in achieving a resolution of the settler’s claimed liability at a relatively early point in the litigation (April 21, 1993 Recommendation of the Special Master at 9-10).
Notably absent from the Special Master’s report, however, is any reference to the governments’ estimates — preliminary or otherwise — of the potential total natural resource damages. Rather, the Special Master’s report simply notes that one of the factors reflective of the consent decree’s reasonableness is its relationship to the “estimates of restoration and response costs.” Nor did the Special Master provide the district court with the governments’ estimates of total damages at the hearing on the motion for approval of the consent decree.3
4. The District Court’s Approval of the LACSD Consent Decree
In its earlier order approving the Pot-latch-Simpson consent decree, the district court, quoting Cannons, observed that “ ‘[i]n settlement negotiations, particularly in the early phases of environmental litigation, precise data relevant to determining the total extent of harm caused and the role of each PRP [potentially responsible party] is often unavailable.’” Montrose I, 793 F.Supp. at 240 (quoting U.S. v. Cannons Engineering Corp., 899 F.2d at 88). Consistent with this observation, the court, in approving the LACSD consent decree, concluded that requiring the parties to provide precise information about the extent of the total damages and the relative culpability of the various defendants would contravene CERCLA’s primary goal of encouraging early settlements. *746United States v. Montrose Chemical Corp. of California, 827 F.Supp. 1453, 1458 (C.D.Cal.1993) (Montrose II).
The district court acknowledged its obligation to “scrutinize” the settlement process to determine whether the decree was both (1) the product of a procedurally fair process, and (2) substantively fair to the parties in light of a reasonable reading of the facts. Id. Acknowledging that obligation and fulfilling it, however, are two different things, and we do not believe that the court could possibly have adequately determined that the settlement was substantively fair without having some benchmark with which to compare it. Apparently, the most appropriate benchmark at the time—the governments’ preliminary damage estimate prepared for purposes of settlement negotiations—was known only to the Special Master, who did not share that information with the court.
Thus, in evaluating the substantive fairness of the consent decree, the court was limited to the observation that the $45.7 million settlement figure “appears to be reasonable and fair” because the parties did not act arbitrarily in arriving at that figure. Noting that the governments had explained “in detail” the methodology used to arrive at the figure, the district court concluded that it would not interfere with their determinations as long as they appeared to be reasonable, especially given that the Special Master supported those determinations. Id.
Relying then on the Special Master’s recommendations, the district court concluded that “the process of settlement and the settlement itself were and are fair and reasonable,” and approved the proposed consent decree.
DISCUSSION
We exercise considerable restraint in reviewing a district court approval of a CERCLA settlement. As the First Circuit recently noted, the battle over whether a consent decree is fair, reasonable and consistent with CERCLA “will usually be won or lost in the trial court.” U.S. v. Charles George Trucking, 34 F.3d 1081, 1084-85 (1st Cir.1994). This is true because a district court’s approval of a CERCLA consent decree reaches the appellate level “encased in a double layer of swaddling.” U.S. v. Cannons Engineering Corp., 899 F.2d 79, 84 (1st Cir.1990); accord In re Cuyahoga Equipment Corp., 980 F.2d 110, 118 (2d Cir.1992) (appellate court owes a “two-fold deference to the approved settlement”).
First, CERCLA’s policy of encouraging early settlements is strengthened when a government agency charged with protecting the public interest “has pulled the laboring oar in constructing the proposed settlement.” Cannons, 899 F.2d at 84. Although “the true measure of the deference due depends on the persuasive power of the agency’s proposal and rationale,” a district court reviewing a proposed consent decree “must refrain from second-guessing the Executive Branch.” Id. Such deference is appropriate given “[t]hat so many affected parties, themselves knowledgeable and represented by experienced lawyers, have hammered out an agreement at arm’s length and advocate its embodiment in a judicial decree.” Id.
The “second layer of swaddling” surrounding a district court’s approval of a CERCLA consent decree is the abuse of discretion standard applied by the appellate court in reviewing the district court’s ruling. Cannons, 899 F.2d at 84; see also Officers For Justice v. Civil Service Com’n., 688 F.2d 615 (9th Cir.1982), cert. denied sub nom. Byrd v. Civil Services Com’n. 459 U.S. 1217, 103 S.Ct. 1219, 75 L.Ed.2d 456 (1983) (court applies abuse of discretion standard to district court’s decision to accept or deny a settlement). Under the abuse of discretion standard, we cannot reverse unless we are convinced that the court below committed a clear error of judgment by failing to apply the correct law or by resting its decision on a clearly erroneous finding of material fact. Marchand v. Mercy Medical Center, 22 F.3d 933, 936 (9th Cir.1994).
Notwithstanding this “swaddling,” there is a fundamental difference in the review of the sufficiency of evidence to support a settlement and the situation where there is no evidence at all on an important point. There is, in fact, no evidence on this record from which the district court could have *747made any determination with respect to estimates of responsibility and damage. It is clear a district court may abuse its discretion by having no evidence to support its decision. MGIC Indem. Corp. v. Moore, 952 F.2d 1120, 1122 (9th Cir.1991).
The First Circuit recently noted in U.S. v. Charles George Trucking, Inc. that the proper way to gauge the adequacy of settlement amounts to be paid by settling PRPs is to compare the proportion of total projected costs to be paid by the settlors unth the proportion of liability attributable to them, and then to factor into the equation any reasonable discounts for litigation risks, time savings, and the like that may be justified. 34 F.3d at 1087 (emphasis added).
Of course, to determine the “proportion of total projected costs to be paid by the set-tlors,” the court must know the total projected costs themselves. Thus, for example, in Charles George Trucking, the court observed that the payment to be tendered by the settling parties, approximately $36 million, represented “more than half of the highest estimate of aggregate clean-up costs ($70,-000,000).” Id. Because the settling parties collectively were responsible for fifty percent of the environmental damage, the court concluded that the settlement was favorable even before considering the various discounting factors. Id.
Here, with no evidence of the governments’ estimate — preliminary or otherwise— of total natural resource damages, the court determined that $45.7 million represented a “fair and reasonable” settlement. However, “fair” and “reasonable” are, by their very nature, comparative terms. In such an informational vacuum, the fairness or reasonableness of a $45.7 million settlement simply cannot be measured.
The governments argue that the district court compensated for its lack of information concerning the overall damage framework by relying on the recommendations of the Special Master, who vouched for the settlement’s fairness. The district court was entitled to rely upon the master, but reliance cannot be so complete that it takes the place of the court’s obligation to independently “scrutinize” the terms of a settlement. See, e.g., U.S. v. Rohm & Haas Co., 721 F.Supp. 666, 680 (D.N.J.1989) (quoting City of Detroit v. Grinnell Corp., 495 F.2d 448, 462 (2d Cir.1974)) (district court “must eschew any rubber stamp approval in favor of an independent evaluation”).4
We therefore vacate the district court’s approval of the consent decree and remand the matter for the district court to conduct an independent evaluation of the settlement with LACSD and the 150 local governmental agencies to determine whether it is “reasonable, fair, and consistent with the purposes that CERCLA is intended to serve.” See Cannons, 899 F.2d at 85.
In conducting that evaluation, the court, in addition to considering any other relevant factors, should determine the proportional relationship between the $45.7 million to be paid by the settling defendants and the governments’ current estimate of total potential damages. The court should evaluate the fairness of that proportional relationship in light of the degree of liability attributable to the settling defendants. See Charles George Trucking, 34 F.3d at 1087.
Moreover, we believe that the nature of the liability of the various defendants is of considerable relevance in determining whether the settlement is fair, reasonable and consistent with the public interest. For example, if joint and several liability does not apply to the natural resources damages, the governments’ ability to collect the totality of remaining damages from the non-settling defendants certainly would have an impact on the settlement’s merits.
Finally, the court should “factor into the equation any reasonable discounts for litigation risks, time savings, and the like that may be justified.”
We recognize that, in holding that the district court erred in approving the proposed consent decree, we tread relatively uncharted *748territory in the area of Superfund litigation. By so holding, however, we do not denigrate CERCLA’s primary goal of encouraging early settlement. Nor do we vitiate the district courts’ considerable discretion in approving such settlements. Deference, however, does not mean turning a blind eye to an empty record on a critical aspect of settlement evaluation. Where clear error occurs, we will reverse. Swaddling is not armor. Because we believe that the district court did not possess sufficient information to adequately determine whether the LACSD consent decree was fair, reasonable, and consistent with CERCLA’s objectives, we vacate and remand.
. The Comprehensive Environmental Response, Compensation, and Liability Act. 42 U.S.C. § 9601 to 9657.
. In addition to Montrose Chemical Corporation of California, the following corporations were identified by the governments as defendant owners or operators of the Montrose DDT plant: Chris-Craft Industries, Inc.; Rhone-Poulenc Basic Chemicals; ICI American Holdings, Inc.; Stauffer Management; and Atkemix Thirty-Seven. These additional corporations are also appellants in this action, and for purposes of this discussion, are referred to collectively as "Mont-rose.”
. The Special Master confirmed only that the governments in fact did have an estimate of damages. He did not volunteer, nor did the court inquire into, the substance of that estimate.
. Appellants argue that the reliance on the master was, in fact, an abdication of judicial responsibility. While we agree that a district court should carefully review a master’s findings to satisfy itself that the standards of CERCLA are met, we find no abdication here.