We report four experiments examining effects of instance similarity on the application of simple explicit rules. We found effects of similarity to illustrative exemplars in error patterns and reaction times. These effects arose even though participants were given perfectly predictive rules, the similarity manipulation depended entirely on rule-irrelevant features, and attention to exemplar similarity was detrimental to task performance. Comparison of results across studies suggests that the effects are mandatory, non-strategic and not subject to conscious control, and as a result, should be pervasive throughout categorization. (Contains 2 figures and 5 tables.)