The purpose of this study is to propose a new effective method to synthesize a natural human body motion in virtual reality. The following experimental studies were conducted especially about human walk motion. (1) Actual human walk motions were measured by using a 3D motion capturing system, and the features of the joints motion were analyzed from the correlations of joint angle transients. (2) The factors which represent the naturalness of human body motion were investigated by 10 subjects'questionnaires after showing various walk motions. As the results of the former study, for example, we found shoulder joints motion and waist twisting motion are similar to each other. As the results of the latter study, we found leg's motion is much more important for human walk motion than the motion of upper half of the body. Then, a new synthesis method has been proposed from the results of the above experimental studies.