3D motion capture data can now be obtained from human body directly. They reflect the real motion of human beings. In this paper, against traditional 2D recognition methods, we propose a novel method for human motion identification from 3D motion captured data. We take a data-driven modeling approach to learn characteristics from a marker-based training set. Principal Components Analysis (PCA) is used to get the low-dimension features from a captured movement sequence. A similarity computing technique is used to compute the distance of different motion clips. The method is tested on motion clips from the CMU motion capture database. All the motions can be identified from the motion clips. The results demonstrate that our method has good identification abilities. Moreover, for periodical motions, the identification results are nearly irrespectively with the length of captured sequence.