In everyday life, we often generalize previously learned associations between our actions and their outcomes to novel, but similar situations. The present study aimed to conceptually and directly replicate results reported by Esser et al. (2023, Psychological Research), who reported that action–effect (A–E) learning generalizes to categorically similar stimuli. Yet, the overall picture in the published literature is not overwhelming with regard to such generalization, and hence a replication seems to be of particular importance. We present three experiments. In our conceptual replications (Exp. 1 and 2), we did not find significant A–E compatibility effects for either learned or similar stimuli. In the direct replication (Exp. 3), we observed evidence for successful A–E learning, but still failed to replicate the generalization effect for similar stimuli. A critical difference between Experiments 1 and 2 and Experiment 3, however, is that stimulus-stimulus (S-S) learning was not possible in the former two experiments, but in the latter one (as was already acknowledged by Esser et al.). In sum, our results (1) suggest a critical role of S-S learning in this particular design (2) and question whether A–E learning truly generalizes to similar stimuli.