Recent studies in development psychology suggest that early on infants are able to distinguish characters who display a cooperative behavior from characters who display an antisocial behavior. The current research builds on these findings and aims at determining the extent to which infants possess the sociomoral distinction of “good” and “mean” agents. In particular, we propose that infants represent sociomoral behaviors through kind-based categories. This hypothesis was tested in the current research across 5 different experiments by investigating how infants represent the identity of agents in sociomoral situations. Experiment 1 used a looking-time paradigm to demonstrate 11-month-old infants’ bias to individuate distinct agents based upon their “mean” and “nice” behaviors in a spatiotemporal ambiguous situation. Experiment 2 and 3 ruled out alternative explanations of this effect by controlling for the number of actions presented and differences in motion, respectively. These findings suggest that infants expect agents to display coherent sociomoral behaviors over time in a particular context. Experiment 4 tested whether infants’ are biased to identify prosocial agents more by their internal than their external properties. Fourteen-month-olds showed a bias to identify the ‘helper’ character based on the color of its internal properties. Experiment 5 aimed to clarify whether infants have this biased because they attribute a causal role to the agents’ internal properties. In two different conditions the causal relevance of the agents’ internal or external property was manipulated. We hypothesized that when the causal relevance of the internal property was undermined infants would no longer be biased toward the agents’ internal properties when identifying it as the “helper” character. So far, the results do not show a clear support for this hypothesis. Overall, the results of all these experiments indicate that infants represent sociomoral behaviors in a relatively categorical fashion, and more strongly associated to the agents’ internal rather than external properties. These findings are discussed from both a strong version of kind-based representations in terms of intrinsic natural kinds, and a weaker version in terms of more graded and extrinsic sociomoral kinds.