We discuss the rather special status of the postulate of "recovery" among the six basic postulates for theory contraction as formulated and studied by Alchourr6n, Gdirdenfors and the author. Our consideration of this postulate brings to the surface an important philosophical issue: whether the objects to which the operations of contraction and revision are applied should be taken to be theories themselves, in the sense of sets of formulas already closed under logical consequence, or bases for these theories. In Section I we note several respects in which the postulate of recovery is different in character from its companions, and perhaps more vulnerable, as well as several respects in which it is, nevertheless, attractive. In Section II we present a small but significant formal result; for every operation on a theory A satisfying the other five postulates for contraction, there is a unique operation satisfying also recovery that is "revision-equivalent" to the former; this unique operation is moreover the greatest among all the operations revision-equivalent to the first and satisfying the other five postulates. Thus in so far as one is interested only in revision, the postulate of recovery is innocuous, facilitating proofs without generating new properties. However, if one is interested in the notion of contraction itself, the question of the epistemological status of the postulate of recovery remains, and this is discussed at length in Section III. We contrast the approach of Gairdenfors, who accepts the postulate from the standpoint of a conception of contraction as a minimal incision in a theory needed to get rid of a proposition, with the approach of Levi, who rejects the idea of minimal loss of content in view of the eccentric formal behaviour to which it gives rise, and rejects recovery along with it. Both of these approaches share the feature that contraction