Abstract: This paper presents a novel approach for retrieving similar motions from large-scale motion capture database with human pose code.Our approach employs nine major bones to represent human pose.First,each bone's active space is partitioned into several subspaces,whose positions are used as feature representations of human poses.Then,a motion data is encoded frame by frame with the feature representation,and consecutive frames which share a motion code are grouped.Thereafter,grouped frames are assigned an inverted index for retrieving representative motions.Similar motions are extracted from these representative motions by code matching algorithm.Experimental results show that our approach can effectively retrieve the similar motions from motion capture database.