Abstract: A content-based motion retrieval approach based on key-frames is proposed.For each motion clip in motion database,the bone angles are employed as motion features and key-frames can be selected based on them.When retrieval command is issued,the distance matrix is constructed between key-frame set of query example and that of a motion clip from database.Then the similarity between them is calculated based on this distance matrix.To validate this retrieval approach,we test it on a database containing about four hundreds motion clips which can be divided into various kinds of human movements.The experimental results demonstrate the effectiveness of our approach.