Stereotype formation about novel groups was analyzed with trivariate stimulus distributions that were generated by group membership, valence of behavior, and a context variable. Within this stimulus setting, we manipulated the confounding role of the context variable and the distinctiveness of events in terms of their relative infrequency. The experimental procedure allowed us to analyze illusory and spurious correlations in a joint framework, to conduct focused tests for memory effects of relative infrequency and to investigate the detection of covariations with the context variable. The results revealed that illusory and spurious correlations were formed without enhanced memory for infrequent events and with existing covariations of the confounding context factor being well extracted. These observations suggest that illusory and spurious correlations can be understood without assuming specific cognitive processes that are tied to the particular characteristics of a given stimulus distribution, such as enhanced memory in the case of relative infrequency and neglect of a context variable in the case of a confounding factor. Instead, computer simulations with an exemplar-based learning model demonstrated that exemplar-based category learning may provide a coherent and integrative theoretical framework for illusory correlations, spurious correlations and true contingency learning in social cognition. Copyright © 2006 John Wiley & Sons, Ltd.