Learning by imitation is a powerful form of learning. Different forms of imitation, like mimicry, copying, response facilitation, etc. have been studied extensively. Recent research in robotics has begun to explore imitation as a means to allow complex robots, like humanoid robots, acquire new skills. One of the key issues in imitation learning is the correspondence problem. This problem concerns the answer to the question: what action sequence of the imitator is similar to that of the demonstrator and how similar it is? The notion of “similarity” has remained subjective thus far. Robotics research in imitation has mostly focussed on action learning and classification, and not on the correspondence problem. Our aim is to develop a generalized metric that provides a scalar measure of dissimilarity/distance between any given pair of action sequences. This, we expect would be a uniform means to evaluate imitation in agents. The metric can also be used as a part of the action selection mechanism in an imitator agent.