To improve the efficiency of information retrieval in P2P networks, there have been many researches on categorizing data items and clustering peers. In almost all these researches, the number of categories and the policy of categorization are predetermined and static. However, users’ requirements for information retrieval dynamically change. This leads to undesired increase of network traffic. In this paper, we propose a dynamic cluster construction method based on query characteristics and a search method using dynamic cluster. Our method dynamically constructs clusters, when the access frequencies for certain data items increase. This approach can reduce the number of query messages for searching data items further than static clustering methods.