Levine's theory of concept identification suggests that more extensive processing follows errors than successes. The role of memory for the stimulus in this processing was studied by asking 40 practiced college students to solve 16 simultaneousdiscrimination problems, each of which had eight dimensions, one of them relevant. Hypotheses were monitored by Levine's blank-trials technique, and each subject was run in one of four conditions depending on whether the stimulus (the positive member of the display pair) was or was not (re-)presented after successes or errors (during success or error feedback). Stimulus presentation after errors increased overall learning rate, global consistency, and local consistency; after successes, it had no effect. Hence, memory for the stimulus seems crucial to the revision of hypotheses after errors. This study is concerned with some specific determinants of the differential effects of errors and successes in concept identification. In conventional concept-identification paradigms, the information given in a sequence of trials does not depend on the responses on those trials. However, most viable theories of concept identification hold that subjects lose more information on errors than on successes. At one extreme, Restle (1962) proposes that after each error the subject forms a set of hypotheses according to a random process that is independent of the preceding sequence of events. Hence, no information given before a particular error will influence behavior after that error. Trabasso and Bower (1968) in a slightly different version of the same theory suggest that after each error the subject samples only locally consistent hypotheses, that is, hypotheses consistent with the information given on only that trial. Hence, the information from exactly one trial, the trial on which the subject made the error, will influence behavior after that error.