Recent behavioral, neuroimaging, and neurophysiological research suggests a common representational code mediating the observation and execution of actions; yet, the nature of this representational code is not well understood. The authors address this question by investigating (a) whether this observation-execution matching system (or mirror system) codes both the constituent movements of an action as well as its goal and (b) how such sensitivity is influenced by top-down effects of instructions. The authors tested the automatic imitation of observed finger actions while manipulating whether the movements were biomechanically possible or impossible, but holding the goal constant. When no mention was made of this difference (Experiment 1), comparable automatic imitation was elicited from possible and impossible actions, suggesting that the actions had been coded at the level of the goal. When attention was drawn to this difference (Experiment 2), however, only possible movements elicited automatic imitation. This sensitivity was specific to imitation, not affecting spatial stimulus-response compatibility (Experiment 3). These results suggest that automatic imitation is modulated by top-down influences, coding actions in terms of both movements and goals depending on the focus of attention. (Contains 5 figures and 4 footnotes.)