This paper describes a method to cope with denormalized relational schemas in a database reverse engineering process. We propose two main steps to irnprove the understanding of data semantics. Firstly we extract inclusion dependencies by analyzing the equijoin queries embedded in application progranis and hy querying the database extension. Secondly we show how to discover only functional dependencies uhicli inthe database extension, the application programs, but especially expert users. * Expressing the extracted semantics with a high level data model This task consists in a schema translation activity and gives rise to several difficulties since the concepts of the original model do not overlap those of the target model. fluence the way attributes should be restructured. The method is interactive since an expert user has to validate the presumptions on the elicited dependencies. Moreover, a restructuring phase leads to a relational schema an third normal form provided with key const...