We examined the effect of mimicry on how 16-month-old infants learn by observation a novel tool use action, which consisted of using a rake to retrieve a toy. Across four conditions, we manipulated whether during an initial play phase, an adult mimicked the infant's play or not (testing the effect of mimicry), the infant played with the adult or played alone (controlling the effect of interacting with a contingent partner) and whether the infant saw a demonstration of the tool's use or not (evaluating baseline performance). We found that infants who had been mimicked learned best from a demonstration of the rake's use and performed better than infants who only played with the experimenter without mimicry or played by themselves before the demonstration. As expected, infants did not learn from a demonstration of the rake's use when they played by themselves and thus had no previous interaction with an experimenter. The mechanisms driving this powerful learning effect of mimicry are discussed.