The affect-as-information hypothesis (e.g., Schwarz & Clore, 2003), predicts that the positive or negative valence of our mood differentially affects our processing of the details of the environment. However, this hypothesis has only been tested with mood induction procedures and fairly complex cognitive tasks in humans. Here, six baboons (Papio papio) living in a social group had free access to a computerized visual search task on which they were over-trained. Trials that immediately followed a spontaneously expressed emotional behavior were analyzed, ruling out possible biases due to induction procedures. RTs following negatively valenced behaviors are slower than those following neutral and positively valenced behaviors, respectively. Thus, moods affect the performance of nonhuman primates tested in highly automatized tasks, as it does in humans during tasks with much higher cognitive demands. These findings reveal a presumably universal and adaptive mechanism by which moods influence performance in various ecological contexts.