The continuous improvement of a Web site's content, can be the key to attract new customers or maintain the existing ones. A way to obtain such improvement, is to study the behavior of a user while browsing in the Web. For the analysis of this behavior two variables are of particular interest: the pages visited during a user session and the time spent in each one of them. The respective Web log files contain part of this data. These files, however, can contain a huge number of registers where large part of them possibly do not contain relevant information. This is one of the reasons why finding initially unknown and useful relations in Web log registers is a complex task, which can be performed applying the process of knowledge discovery in databases (KDD). We propose a methodology for Web mining based on a data mart model. We applied this methodology analyzing log files from a certain Web site. The respective results, gave very important insights regarding visitors behavior and preferences. This knowledge has been used in the Web site's reconfiguration.