Four experiments were designed to evaluate the functional correspondence of effective performance with correct or incorrect instructions and correct or incorrect self-descriptions in a first-order matching-to-sample task. These studies included verbal or nonverbal matching responses and provided feedback or not after the participants described their matching performance. The results point to three possible discrimination learning processes in humans: (1) learning through instructions, with a possible insensitivity to consequences unless the correspondence between instructions and feedback breaks down; (2) learning through feedback, with an inability of participants to describe their own behavior; and (3) a genuine "rule-governed" behavior consisting of successful task performance and explicit verbal behavior describing the actual contingencies effective for such performance.Matching to sample is the most frequently used task in the study of conditional discrimination learning in animal and human participants (Goldiamond, 1966). In conditional discrimination, the particular properties of the events that qualify as discriminative stimuli (SOs) or as S-delta stimuli (S~s) usually vary according to certain relational criteria (transposition, matching). In the case of matching-to-sample procedures, the relevant properties of SOs and S~s may be related to a single sample stimulus (first-order matching to sample), or to a relation between two or more stimuli (second-order stimuli) that are added to the first-order task (second-order matching to sample).1 The relational criteria most widely used in matching to sample have been identity (when stimuli share all of their properties), similarity (when stimuli share only 1 In second-order matching-to-sample, two second-order stimuli, one sample stimulus, and at least two comparison stimuli are presented. The second-order stimuli specify the "correct" criterion for matching the sample and comparison stimuli. Thus, if the two secondorder stimuli are identical in color and shape, the correct comparison is the one matching the sample in color and shape.We acknowledge the suggestions made by Fran,