Human recognition by gaits is a new biometric method which considers both spatial and temporal features by tracking the gait at a distance. Most researches use image processing methods to analyze the moving object, while some use the model-based methods. Although some efforts have been devoted to the gait recognition, a high identification rate and a low computational cost still need to be achieved. This paper tends to investigate the gait recognition from the biomechanical point of view, which is lacking in the computer science and electronics communities. We present the method of people recognition by using both three-dimensional parameters of lower limb joints, namely, kinematic parameters (including joint angles and angular velocities) and kinetic parameters (joint forces). Dynamic data are acquired by tracking the maker positions using a motion capture system. 350 trials for 10 subjects are conducted, in which we use 20 trials for each one of these for training and the remaining 15 for testing. Additional 35 trials for another one subject not in the training set are conducted for testing the adventitious people. SOM neuron networks are employed for data classification. The importance of specific joints and variables is discussed in detail to investigate the major factors that cause the differences in human gait through a biomechanical approach. Experimental results show that gait is a reliable feature for individual identification since a high recognition rate can be achieved by choosing the proper joints or dynamic parameters of a gait. It also suggests that some joints and dynamic parameters dominate the differences in the walking style of people, which can be applied to both the biomedical and the computer vision fields.