By grouping similar Web documents into clusters, the search space can be reduced, the search accelerated, and its precision improved. In this paper, a new clustering algorithm is introduced. In the clustering technique, topics are represented according to VSM (vector space model), documents are represented according to topics, and the relation between documents and topics is viewed in a transactional form, each document corresponds to a transaction and each topic corresponds to an item. A frequent item sets can be found by using the association riles discovery algorithm,corresponding documents can be seen as initial clusters.These clusters are merged according to the disance between clusters,or divided aivided according to the strength of connection among documents of a cluster.By real Wed documents,experimental results show the algorithm's effectivenss and suitability for tackling the overlapping clusters inhered by documents.