concurring.
I concur in the judgment and in all but Part III of the court’s opinion. Like the opinion of the district court, Part III is too broad, in my view; I would allow recovery on a narrower ground.
The removal of waste-laden drums and contaminated soil from the Stevens Landfill constituted a single removal action, as I see it, notwithstanding that some of the material was removed by one contractor and some by another. Under the applicable statute of limitations, 42 U.S.C. § 9613(g)(2)(A), a cost-recovery suit may be brought within three years after completion of a removal action. Because this particular removal action was not completed until July 15, 1987, and because the state of Michigan commenced its lawsuit on July 12, 1990, the statute does not bar recovery of any of the costs associated with the removal action.
*846This conclusion would be inescapable, it seems to me, regardless of whether a Remedial Investigation/Feasibility Study (“RI/FS”) had been conducted at the site and regardless of the timing of any such investigation/study. Under the interpretation that the United States Environmental Protection Agency placed on the statute of limitations within a year of the statute’s enactment, a Remedial Investigation/Feasibility Study “should be treated as a separate removal action.”1
A Remedial Investigation/Feasibility Study was conducted at the Stevens Landfill, of course, and it was not completed until November of 1987. The instant lawsuit having been filed on July 12, 1990, the defendants concede that the three-year statute of limitations does not bar recovery of the costs associated with the RI/FS. Indeed, as the defendants’ brief tells us, all of such costs have been paid.
In a notice of proposed rulemaking published in August of 1992 — almost six years after the enactment of 42 U.S.C. § 9613(g)— the U.S. EPA proposed a regulatory amendment reading, in part, as follows:
“For purposes of defining certain events that determine the statutory limitation periods applicable to cost-recovery actions pursuant to section 113(g)(2) of CERCLA [42 U.S.C. § 9613(g)(2) ], the following applies:
a. The term ‘completion of the removal action’ for sites where remedial actions are taken means the date of the final remedial design prepared in connection with the final remedial action at the site.” 57 Fed. Reg. 34742, 34754 (August 6, 1992).
The preamble accompanying the proposed rule explains that there are different kinds of removal actions (“traditional physical removals,” e.g., as distinguished from “studies or investigations conducted under section 104(b) of CERCLA”), and the preamble notes that because “several of these removal actions may be taken simultaneously or in sequence at a site, the completion of the removal action for purposes of cost recovery may be difficult to ascertain.” 57 Fed.Reg. at 34751. To circumvent this difficulty, the agency proposes to treat all removal actions at a given site as a single removal action that will be deemed incomplete until “the date of the last remedial design report prepared by EPA preparatory to implementation of remedial construction activities at the site.” Id.
Whether the agency will actually adopt such a rule is unknown at this point. One factor militating against adoption, I should think, is the obvious difficulty of reconciling the language of the proposed rule with the language of the statute. To recover costs for “a removal action,” as the statute says in the plainest of English, the plaintiff must commence suit within three years after “the removal action” has been completed. This means — if words have meaning — that suit must be commenced within three years after completion of the individual removal action the costs of which are sought to be recovered, not three years after the date of a remedial design report that may not be issued until long after completion of the removal action.
It bears emphasis that the preamble to the rule proposed in 1992 makes no attempt to back away from the proposition that the cleanup of a particular site may involve multiple removal actions. In addition to the material already referred to at 57 Fed.Reg. at 34751, see 57 Fed.Reg. at 34743:
“In order for EPA to undertake [‘traditional’ physical] removal actions and pay for them out of the Superfund, the action must cost less than $2 million and last less than one year, unless a waiver is granted pursuant to section 104(c)(1) of CERCLA This statutory limitation does not apply to removal actions taken by responsible parties.
In addition to these ‘traditional’ physical removals, CERCLA’s definition of removal *847also includes certain other activities, including studies and investigations of releases and threats of releases to determine their nature and extent, and other actions taken to plan and direct response actions pursuant to section 104(b) of CERCLA.”
Yet under the proposed rule, a $2 million physical removal action completed in 1987, for example, could be the subject of a cost-recovery suit commenced in 1992 or 1993 if the agency simply delayed long enough in issuing a final remedial design report. Such a result would be administratively convenient, no doubt, but I see nothing in the statute that could reasonably be construed as authorizing the agency to treat as incomplete a physical removal action that was, in fact, complete.
If the agency had read the statute this broadly from the beginning, I suppose the agency’s interpretation might be entitled to some measure of deference. Administrative practice has “peculiar weight,” as Mr. Justice Cardozo once said, “when it involves a contemporaneous construction of a statute by the men charged with the responsibility of setting its machinery in motion, of making the parts work efficiently and smoothly while they are yet untried and new.” Norwegian Nitrogen Products Co. v. United States, 288 U.S. 294, 315, 53 S.Ct. 350, 358, 77 L.Ed. 796 (1933) (emphasis supplied).
When the 1986 statute was still “untried and new,” however, the contemporaneous interpretation given it by EPA was diametrically opposed to the interpretation reflected in the 1992 proposed rule that the agency is now thinking about adopting. The latter interpretation is not final, it is not contemporaneous, and it is not compatible with the language of the statute. Accordingly, in my opinion, the rule of deference does not require us to follow it. See Standard Oil Co. v. Dept. of Energy, 596 F.2d 1029, 1056 (Em.App.1978):
“Deference to an agency’s ‘interpretation’ ... is not a hard and fast rule. The weight to be given to an administrative interpretation depends upon ‘the thoroughness evident in its consideration, the validity of its reasoning, its consistency with earlier and later pronouncements, and all of those factors which give it power to persuade, if lacking power to control.’ Skidmore v. Swift & Co., 323 U.S. 134, 140, 65 S.Ct. 161, 164, 89 L.Ed. 124, 129 (1944).”
“An agency interpretation of a relevant provision which conflicts with the agency’s earlier interpretation is ‘entitled to considerably less deference’ than a consistently held agency view.” INS v. Cardoza-Fonseca, 480 U.S. 421, 446 n. 30, 107 S.Ct. 1207, 1221 n. 30, 94 L.Ed.2d 434 (1987), citing Watt v. Alaska, 451 U.S. 259, 273, 101 S.Ct. 1673, 1681, 68 L.Ed.2d 80 (1981).
I am not persuaded that EPA’s original reading of the statute of limitations was wrong, and I would decide this case in a manner consistent with the agency’s original understanding.