Body motions associated with walking exhibit irregular and complex patterns with time. Chaos analysis methods have been developed to clarify nonlinearity of the lower extremity motions. No research has been reported on chaos analysis of the upper extremity joints. The purpose of this study was to investigate chaotic characteristics of movements of the upper body as well as the lower extremity during level walking. Gait experiments were carried out for eighteen young males. Each subject was instructed to walk on a treadmill at his own natural speed. Flexion-extension angles of eleven joints were obtained by using eight video cameras. To evaluate joint characteristics in a quantitative way, the largest Lyapunov exponent(LLE) was calculated from a reconstructed state space created by time series and embedding dimension. The mean LLE ranged from 0.080 to 0.137 for the upper body, and from 0.090 to 0.182 for the lower extremity joints. The mean LLE of the upper extremity joints was statistically different from that of the lower extremity joints(p<0.05). The results obtained can be used as a valuable reference for the normal gait for further studies of abnormal walking.