The transmission of cultural knowledge requires learners to identify what relevant information to retain and selectively imitate when observing others' skills. Young human infants--without relying on language or theory of mind--already show evidence of this ability. If, for example, in a communicative context, a model demonstrates a head action instead of a more efficient hand action, infants imitate the head action only if the demonstrator had no good reason to do so, suggesting that their imitation is a selective, interpretative process [1]. Early sensitivity to ostensive-communicative cues and to the efficiency of goal-directed actions is thought to be a crucial prerequisite for such relevance-guided selective imitation [2]. Although this competence is thought to be human specific [2], here we show an analog capacity in the dog. In our experiment, subjects watched a demonstrator dog pulling a rod with the paw instead of the preferred mouth action. In the first group, using the "inefficient" action was justified by the model's carrying of a ball in her mouth, whereas in the second group, no constraints could explain the demonstrator's choice. In the first trial after observation, dogs imitated the nonpreferred action only in the second group. Consequently, dogs, like children, demonstrated inferential selective imitation.