This paper reports our contribution to the POESIA project, aimed at the automatic filtering of illicit contents on the Internet. In this context, we present an original text classification system, based on information theory principles, which is able to detect and filter with a high accuracy pornographic texts