Abstract: An improved K-means clustering algorithm based on hierarchical micro-clustering(HMKC) was proposed, combining the hierarchical structure of space, and the criterion function S(k) was reconstructed. Firstly, a hierarchical K-means clustering tree was produced by using hierarchical clustering algorithm, then the center points were updated dynamically on the tree structure according to the micro-clustering method. Finally, the improved criterion function S(k) was used to find rational clustering number K and corresponding core point to ensure the clustering result has reached optimal globally. Experimental results on standard datasets demonstrate that the effect and accuracy of clustering results can be improved significantly with the HMKC algorithm comparing with the traditional k-means algorithms.