Although children can use social categories to intelligently select informants, children's preference for in-group informants has not been consistently demonstrated across age and context. This research clarifies the extent to which children use social categories to guide learning by presenting participants with a live or video-recorded action demonstration by a linguistic in-group and/or out-group model. Participants' (N = 104) propensity to imitate these actions was assessed. Nineteen-month-olds did not selectively imitate the actions of the in-group model in live contexts, though in-group preferences were found after watching the demonstration on video. Three-year-olds selectively imitated the actions demonstrated by the in-group member regardless of context. These results indicate that in-group preferences have a more nuanced effect on social learning than previous research has indicated.