Abstract: In order to find Web users with similar access interest, clustering of Web users is studied in this paper. Access records of the Web users are extracted from Web severs’ log files and organized into user sessions. Each session is a compact sequence of Web accesses by a user. Using attributed oriented induction, the sessions are then generalized according to the page hierarchy. A new distance is defined to measure similarity between two generalized sessions. The generalized sessions are finally clustered using some non Euclidean and relational clustering algorithms. The experiment shows that this approach is efficient and practical in finding several interesting clusters within a large set of log.