Schema matching is a crucial task to gather information of the same domain. However, this process is still largely performed manually or semi-automatically, discouraging the deployment of large-scale mediation systems. Indeed, these large-scale scenarii need a solution which ensures both an acceptable matching quality and good performance. In this article, we present the BMatch approach to efficiently match a large number of schemas. The quality aspect is based on the combination of terminological and contextual methods. The performance aspect relies on a B-tree indexing structure to reduce the search space. Finally, experiments with real sets of schemas show that our approach is scalable and outperforms the most referenced matching tools both in quality of matches and performance time.