The aim of this project is to propose a methodology in automatically building thematic index from French medical texts in order to improve the IR process. In this article, we focus on the selection process of relevant terms. Contrary to Bourigault and Charlet (1999) who defend a statistical method followed by human intervention, we propose an automatic method that takes advantage of available a priori medical resources such as the MeSH thesaurus (Lindberg et al., 1993) and GALEN (Rector et al., 1996).