Data integration by mediation is more practical then centralized data warehouse when data is not static. The use of ontologies in the mediation allows semantic and structural integration. In this paper, we propose a new mediation system based on a hybrid architecture of ontologies modeled according to GLAV (Generalized Local As View) model. Thus, we propose an ascending method for building local and global ontologies, which facilitates the semantic reconciliation between data sources. Moreover, we use OWL (Ontology Web Language) for defining mappings between data sources, and ontologies. We also propose a language to formulate user queries. The query language handles global ontology concepts and local ontologies properties because we assume that the user is expert in its domain. Finally, we propose a query rewriting strategy: queries are decomposed to obtain a set of equivalent subqueries that are sent to the corresponding data sources for execution, and after that recomposed to obtain the final result.