A large amount of research has shown that there are two types of trait content in social cognition – agency (including competence) and communion (including morality). Because communal traits are more instrumental in locating a person on the approach-avoidance dimension than agentic ones, the former are considered to be relatively more important in person (and group) perception processes. We developed a proposal that this difference in importance extends to spontaneous trait inferences based on the behavior of the perceived person. The hypothesis that trait inferences are stronger in the communal than agentic domain was tested in four experiments (N = 265) using three different methods of studying spontaneous trait inferences (i.e., the cued recall of distractors procedure, the false recognition paradigm, and the lexical decision task). Despite the variation in methods, the studies yielded the same result – spontaneous trait inferences appeared stronger in the communal than agentic domain, but the effect was restricted to the traits of positive rather than negative valence. For the agentic domain the strength of trait inferences remained relatively low, independent of trait valence. Possible reasons for the difference between positive and negative communal traits are discussed.