The naming latency of a pronoun was measured when a single previously presented name in a discourse either agreed or did not agree with the pronoun in gender and person. An effect of agreement was found both under conditions in which subjects were likely to have engaged in strategic processing of the pronoun (Experiment 1) and under conditions in which this was unlikely (Experiment 3). The effect of gender agreement was also investigated when two noun phrases were present in the discourse. The results continued to show an immediate effect of gender agreement (naming latencies increased when a pronoun did not agree with one of two previously presented nouns) under experimental conditions likely to engender strategic processing (Experiment 2). This last effect was not significant under experimental conditions that were not likely to engender strategic processing (Experiment 3). The results are discussed in terms of models of the process of identifying the referent of a pronoun in a discourse.