Abstract: In recent years,machine learning techniques have been widely used in the synthesis of human motions and many excellent results are achieved,by taking advantage of the mature and the wide use of motion capture.Motion capture allows the data-driven methods become the mainstream techniques for human motion generation.Moreover,machine learning techniques are capable of handling effectively the high dimensionality of human motion data and their correlations in spatial and temporal domain.Based on the characteristics of various machine learning techniques,a survey on the application of machine learning in human motion synthesis is given in this paper.The future development in this research area is also discussed.