Abstract: Motion recognition and simulation are important in human motion analysis. In this paper, an experimental motion analysis system based on retrieval is implemented. Wavelet moments are translation, rotation and scale invariant, and can extract local multi-level features, so they are applied as features to describe motion sequence and actions. Motion recognition is implemented based on the similarity of motion sequences, while action match is addressed using dynamic time warping (DTW). Resulted motion simulation is represented in 3D models. Experiments show the algorithm based on retrieval is efficient for motion recognition and simulation. Retrieval results can simulate human motion, or provide the coarse human motion configuration data for further analysis.