Within the context of two sets of robotic experiments we have performed, we examine some representational and algorithmic issues that need to be addressed in order to equip robots with the capacity to imitate. We suggest that some of the di culties might be eased by placing imitation architectures within a wider social context.