In 5 experiments, participants read passages with sentences containing a categorical anaphoric reference. The exemplar referred to was either present in the passage, present but negated, or absent from the passage. Reading times were at least as fast when the exemplar was absent as when it was present, and reading times were slowest when the exemplar was negated. Naming and recognition probe data from several experiments demonstrated that the exemplar was reactivated in both the exemplar-present and exemplar-negated conditions. The results support a 2-stage process in which a rapid sample of information reactivated by the anaphoric reference determines whether a second, slower, integration stage will be initiated.