Prior categorization studies have shown that pigeons reliably track features that are perfect predictors of category membership (Castro & Wasserman, 2014; 2016a). One might further ask whether pigeons would also track features that are relevant, but imperfect predictors of category membership. In our present project, pigeons had to categorize multiple exemplars from two different artificial categories, in which the exemplars were composed of four different features, that were associated with one of two different report responses. Each exemplar contained one feature that perfectly predicted category membership; one feature that imperfectly predicted category membership; and, two irrelevant features that did not predict category membership. We monitored pigeons’ choice accuracy as well as the location of their pecks to each of the four exemplar features to determine to which attributes the birds attended. As categorization accuracy rose, pecks to the perfect predictor of each category rose as well. Pigeons also showed evidence of attending more to the imperfect predictor than to the irrelevant features, but to a lesser degree. Overall, our results provide evidence of selective attention in pigeons’ categorization behavior.