The Quantitative Kidney DataBase (QKDB) is relational database that was created in order to centralize kidney-related experimental results. Each result is characterized by different attributes and the scientific paper from which it was extracted. Currently, this database isp opulated by hand by experts of the domain. We present a corpus study of some papers that have already been analyzed in order to exhibit the specificities and difficulties of the extraction process; then we propose a first solution to extract automatically the desired knowledge from papers.