The purpose of this study was to delineate Ss' capabilities for learning to recognize different levels of stimulus variability in a schema-formation task. Ss were required to classify, without knowledge of results, 30 sets (6 stimuli per set) of complex, multidimensional histoforms which were distortions of three different population prototypes; Ss then provided scaled judgments of stimulus variability following each classification response. Results showed that Ss whose classifications were consistent with the prototype-defined categories also learned to recognize several different magnitudes of prototype distortion, whereas the inconsistent Ss failed to exhibit any degree of variability learning. The supposition that deviant stimuli may be processed and stored in memory as pattern-specific corrections was thus supported, at least with respect to the variability-learning requirement.