Experiments were conducted to demonstrate the utility of a rule-plus-exception model for extending current esemplar-based views of categorization and recognition memory. According to the model, exemplars that are exceptions to category rules have a special status in memory relative to other olditems. In each experiment, subjects first learned to categorize items organized into two ill-defined categones and then made old-new recognition judgments. Although the categories afforded no perfect rules, a variety of imperfect rules could be formed combined with memorization of exceptions to those rules. In Experiment 1, subjects were given explicit rule-plus-exception instructions. In Experiment 2, subjects were given standard free-strategy instructions and were later clustered on the basis of generalizations consistent with alternative rules. In Experiment 3, a category structure was used that allowed the exceptions to be predicted a priori. In each experiment, differential recognition of exceptions to imperfect logical rules was found, consistent with the predictions of the rule-plus-exception model. In Experiment 4, these findings were extended to prototype-plus-exception structures involving stimuli varying along continuous, fuzzy dimensions. 2