PSYCHOLOGICAL SCIENCE Commentary RECOURSE TO STORED EXEMPLARS IS NOT NECESSARILY EXPLICIT:

learning in normal and amnesic subjects. Amnesics per-formed as well as control subjects on grammaticality judgments, but were impaired on both item recognition and classification when explicit comparison to prior ex-emplars was emphasized. Both of these impairments were attributed to deficiencies in explicit retrieval. Based on these effects, Knowlton et al. concluded that "classification based on rules can occur without recourse to stored exemplars " (p. 177). This conclusion requires that explicit retrieval of items be equivalent to the use of exemplars. However, many contemporary models of classification and recognition memory have a mechanism that retrieves a "chorus of instances " similar to the probe item (e.g., Hintzman, 1986; Kahneman & Miller, 1986; Nosofsky, 1986). This retrieval-time averaging of instance-based knowledge has been provided as an ex-planation of "implicit abstraction " in some classification tasks (Hintzman, 1986) and for implicit, vague feelings of familiarity in recognition memory tasks (Atkinson & Juola, 1974; Jacoby, 1991; Mandler, 1980). Thus, al-though the demonstration of comparable classification learning with normal and explicit-memory-impaired sub-jects is interesting, it has little bearing on exemplar theo-ries of memory and classification becl:luse none posit ex-plicit retrieval as the primary basis of this performance,(see Whittlesea & Dorken, 1993). Furthermore, because 'of retrieval-time averaging and nonlinear similarity func-itions, even the single-factor models have resources to explain the divergences between the two tasks reported by Knowlton et al. (see Hintzman, 1986, and discussion in Vokey & Brooks, 1992). The similarity distribution of 'the test instances has to be controlled and evaluated more tightly than in the study by Knowlton et al. to rule out such models., A second problem is that the multiple response bases of both recognition memory and grammaticality judg