In this paper, a fuzzy rough approximation approach is proposed to cluster user access patterns from Web logs. In the process of clustering user access patterns, whether a Web page is visited or not discloses a Web user's interest during a surfing. Time duration on a Web page reveals the degree of a Web user's interest. And it is characterized as a fuzzy linguistic variable. Each user access pattern is denoted by a fuzzy vector representing visited Web pages and time durations during a surfing. Finally rough approximation approach is adopted to cluster user access patterns denoted by fuzzy vectors. Using this approach, user access patterns with similar surfing behaviors are clustered into one class.