In our project about the digital library (DL) of scientific theses, we need to allow the user an access to the most pertinent information. Therefore, it is important to extract the main concepts to improve the information retrieval in this area. This article represents an empirical evaluation of four tools for automatically extracting concepts from documents. We have compared these tools by using different document collections. For each document, we have extracted manually a list of concepts tied to the main topics. The tools are evaluated according to the degree of similitude between the concepts defined manually and the concepts automatically extracted by these tools. The four evaluated tools are: (1) TerminologyExtractor of Chamblon Systems Inc., (2) Xerox Terminology Suite of Xerox, (3) Nomino of Nomino Technologies, (4) Copernic Summarizer of NRC. This article presents the criteria of evaluation, a comparative study of the tools, an evaluation of the results and a proposed tool to annotate documents based on the concepts extracted.