Most of Web sites are built as a matter of priority. Therefore, to reduce the development time, the conceptualization phase is often put aside and the associated documentation is neglected. Moreover, during the exploitation phase, Web sites suffer the effects of a rapid and unstructured evolution process. Their reconstruction encompasses inevitably a reverse engineering process. In this paper, we propose RetroWeb, a reverse engineering approach of semi-structured Web sites. It aims to provide a description of the site informative content at the physical, logical and conceptual levels. This approach uses, at each level, a meta-model which is instantiated using reverse engineering rules.