Much effort has been made to provide tools for searching information on the Web. As a result, many web search engines are available for user to utilize the rich web resources. However, due to the enormous size of the Web and low precision of user queries, results returned from present web search engines are not always satisfactory. One approach to resolve this problem is by expanding the query using "correlated" terms. In this paper, a method for expansion query based on tolerance rough set is presented, which made use of the term co-occurrence to enrich description of Web query. Experiments on the proposal approach were performed. The precision comparison between the proposal system and the well-known Web search engine --Google on a collection of 5,000 web pages was presented. As a result, improvement of the results precision was achieved by using tolerance classes.