Abstract: Clustering display of search results has been proved an efficient way to organize the Web resources.However，for a given query，clustering results reached by any user are totally identical.A novel search method based on clustering is proposed，which is a modified version of k-means algorithm.The results generated from usual search engine go through a clustering stage based on user interests to create user-specific clusters.Experiments show that it can offer user-specific information needs，improve clustering effectiveness and searching efficiency.