Abstract : Information access systems like Web Sites or Intranet are generally characterized and personalized by using Web Data Mining techniques. The evaluation process is decomposed into 3 steps : first circulation data (stored in log files) are pre-treated in order to reduce noise, then, in function of the aim of the study, analytical methods are choice and applied, and finally results are interpreted. In the first part of this paper, we describe different pre-treatments and present an overview of mathematical and statistical methods used. In a second part, we present, thought our own experiment, the difficulty of interpreting the results. Our experimentation's aim is to discover interesting usage patterns from a pedagogical Web Site data in order to adapt and personalize the presentation. The software used is Clémentine© (SPSS sociéty).