Abstract: Summary Underlying the various behaviors that are classified as imitation, there may be several distinct mechanisms, differing in adaptive function, cognitive basis, and computational power. Experiments reporting “true motor imitation” in animals do not as yet give evidence of production learning by imitation; instead, contextual imitation can explain their data, and this can be explained by a simple mechanism (response facilitation) which matches known neural findings. When imitation serves a function in social mimicry, which applies to a wide range of phenomena from neonatal imitation in humans and great apes to pair-bonding in some bird species, the fidelity of the behavioral match is crucial. Learning of novel behavior can potentially be achieved by matching the outcome of a model's action, and it is argued that vocal imitation by birds is a clear example of this method (which is sometimes called emulation). Alternatively, the behavior itself may be perceived in terms of actions that the observer can perform, and thus it may be copied. If the imitation is linear and stringlike (action level), following the surface form rather than the underlying plan, then its utility for learning new instrumental methods is limited. However, the underlying plan of hierarchically organized behavior is visible in output behavior, in subtle but detectable ways, and imitation could instead be based on this organization (program level), extracted automatically by string parsing. Currently, the most likely candidates for such capacities are all great apes. It is argued that this ability to perceive the underlying plan of action, in addition to allowing highly flexible imitation of novel instrumental methods, may have resulted in the competence to understand the intentions (theory of mind) of others.