Abstract: It has usually ignored the usage of knowledge when querying data in a traditional database. By the rapid development of network, data are distributed to different physical sites. It is almost impossible for the users to know the data wholly in such complicated cases. In this paper, we present a rough set-based query strategy in a distributed knowledge system and concentrate on the role of knowledge. It can reduce the communication costs among the different sites of network, and enhance the query efficiency accordingly. Rough set theory makes the query rough or fuzzy, not being precise as in the conventional case.