Abstracting a motion to a set of features results in a loss in temporal information and variance in the motions is typically treated as noise due to the deterministic nature of

We propose a stochastic framework to analyze and compare differences in human motions for applications in injury prevention, rehabilitation, sports training and performance research. Human motions are modeled as Hidden Markov Models and the differences between the motions are measured using the Kullback-Leibler distance metric. The distance metric is recomputed with degrees of freedom excluded to determine which degree of freedom most influences the difference between a set of motions. The proposed system is tested on a human motion dataset consisting of lifting movements under differing load weights and ankle bracing conditions. Results indicate that the algorithm is capable of successfully determining which joints are impacted and ranking them according to importance.