Abstract: Currently, motion data are often stored in small clips for being used in animations and games. So the behavior segmentation of motion data is a key problem in the process of motion capture. In order to segment the motion data into small clips, a new symbolic representation of motion capture data is introduced and a behavior segmentation approach based on the representation is explored. The high dimensional motion capture data are first mapped on a low dimensional space, based on spectral clustering and sliding-window distance extending weighted quaternion distances. Then the low dimensional data can be represented by a character string, called motion string (MS), and by temporal reverting and max filtering. Because MS converts motion data into a character string, lots of string analysis methods can be used for motion processing. In addition to motion segmentation, motion string may be widely applied in various other areas such as motion retrieval and motion compression. Suffix trees are used to segment the moion data by extracting all static substrings and periodic substrings from MS. Each substring represents a behavior segment, and the motion data are segmented into distinct behavior segments by annotating these substrings. In the experiments, MS is proved to be a powerful concept for motion segmentation, providing the good performance.