Humans routinely monitor social interactions to learn about the relational make-up of their groups and select social partners. It is unclear however whether social interactions primarily invite inferences about the dispositions of the participants involved or about underlying social relations. In the present study we tested which of these two inferences children and adults draw when observing interactions based on dominance. Children expected dominants to prevail over previous subordinates but did not generalize this expectation to interactions with novel agents, whereas adults did. These results suggest that children interpreted dominance as specific to a particular social relation, whereas adults interpreted it as a stable, target-invariant trait. This asymmetry supports the proposal that children may first interpret social interactions through a relational stance, and only later in development apprehend them through the lenses of trait attribution.