Unlike many works in Web Usage Mining, which took as object of study logs recorded by Web servers; we develop in this paper a different approach, qualified as usercentric, judged more accurate and effective. To do so, we develop first a client side tool, in order to collect the user navigation traces. Based on Browser Helper Object, this tool has several advantages as the lightness, the simple exploitation, and almost absence of changes in user environment. Secondly, we elaborate a pre-processing application to prepare the raw client logs. This application includes numerous algorithms and modules, which are originals and appropriates to the log defined format, in order to clean it, reconstruct user sessions and to do some final formatting tasks. The last stage in our work is devoted to the task of client web session clustering using Kohonen Self organizing Maps, with the use of a free knowledge discovery tool. ; Comment: 14 pages, 6 figures