In this paper, we present a cognitive model that simulates the processing of subject pronouns in Italian. The model is implemented in the cognitive architecture ACT-R and uses hierarchically ranked constraints to select the most likely referent of a pronoun. When this model is combined with a measure of accessibility in discourse and a processing time limit imposed by the speed of natural language production, the model accounts for novel empirical data of the interpretation of null as well as overt subject pronouns in Italian. The model generates concrete predictions on the basis of variations in cognitive capacities, which can be tested in subsequent experiments.