Detecting the most probable {it next} page a user is bound to visit inside a website has important practical consequences: it allows to suggest recommendations to the visitors as to which may be the pages of interest to them in a complex website; it is of help for website designers for deciding how to organize the site contents and it is also useful for pre-caching voluminous objects that the user will very probably need. In sum, it helps to customize web contents. In order to achieve that goal a classification, prediction an evaluation cycle has to be performed. Among the several possible alternative technologies we discuss a real use of Bayesian Network representations. The obtained results are commented, compared to other approaches and its applicability to other domains is also discussed.