In two experiments, this paper examines how the labels used to describe interpersonal interactions can affect perceivers' judgments of who caused the interaction. Two universal, connotative dimensions of word meaning underlying the labels, evaluation and potency, influenced expectations about interactants' behaviors and experiences, which in turn affected perceivers' causal attributions. Evaluation and potency ratings for a set of experiencer verbs, a set of action verbs, a set of trait labels (Experiment 1) and a set of social category labels (Experiment 2) were used to construct sentences describing interactions between two people. The complete set of sentences contained all possible combinations of high or low evaluation and potency for all the sentence constituents. Participants were asked to judge who caused the event—subject or object—without having been told that evaluation and potency were the dimensions of interest. When the sentence subject and object differed in evaluation, the evaluative match between the sentence subject and the verb was the most important factor influencing attributions. The potency of the constituents and the class of the verb (experiencer or action) affected the magnitude of the attributions. When the sentence subject and object shared the same valence, attributions were based on verb class. The results highlight the important role of language in interpreting social behavior. Copyright © 2002 John Wiley & Sons, Ltd.