Abstract. This work takes place in the general context of the construction and validation of a domain expertise. It aims at the cooperation of two kinds of knowledge, heterogeneous by their granularity levels and their formalisms: expert statements represented in the conceptual graph model and experimental data represented in the relational model. We propose to automate two stages: firstly, the generation of an ontology (terminological part of the conceptual graph model) guided both by the relational schema and by the data it contains; secondly, the evaluation of the validity of the expert statements within the experimental data, using annotated conceptual graph patterns.