This paper describes the work done in the TIPS project about the construction of a thesaurus base. This construction is a merge from a thesaurus manually built and one automatically extracted from large text corpora. Several manually built thesaurus have been semi-formatted to be merged in a consistent common base. The automatic extraction is based on both syntax and statistics. We present in this paper the way thesaurus are built and the results on Scientific corpus in the context of the TIPS project.