The SNN similarity and density based clustering, as one of the most important unsupervised clustering method, has been proved to produce good results in finding clusters of various sizes and shapes. But these algorithms still have some limitations. For example, Jarvis-Patrick scheme of finding clusters by single link, may separate real clusters or merge clusters which should be kept separated in certain situations, and the determination of Eps and MinPts, the parameters of SNN density method, is hard for users. To deal with these problems, the paper gives analysis and definition of local agglomerative characteristics presented in clustering procedure; then proposes a new clustering algorithm which use local gathering features to control clustering progress. The algorithm can work well in finding different size and density clusters, highlighting the local features of data analysis and improving the quality of data clusters.