The present study investigated young children’s causal generalizations by examining their inductions from second-order learning—where learned correlations between two pairs of features (A–B and A–C) are generalized to non-contiguous features (B–C). Specifically, we asked whether 3- to 6-year-olds could engage in such learning across two canonical causal event types—Blicket detector and Michottian launching—while manipulating contact as a perceptual cue for causality. We replicated Benton, Rakison, and Sobel (2021), finding that children consistently applied second-order learning to infer which objects would produce causal outcomes. Crucially, children’s responses did not differ overall between event types. However, there was a significant interaction by age and task condition: younger children learned better when objects did not contact, whereas older children learned better from contact events. Results are discussed with respect to implications for the development of children’s causal expectations.