Abstract: Many clustering algorithm request users to identify the number of clusters before cluster data.This is very difficult for users.In this paper，clusters which are bigger than intra similarity threshold value are split repeatedly.At last，the clusters which are smaller than inter similarity threshold value are merged to have the final number of clusters.Experiments show that the number of clusters identified by the algorithm is equal to the natural number of clusters，and the intra similarity is high，the inter similarity is low，so the algorithm is easy and efficient.