The emergence of spatial data infrastructures offering geospatial information from heterogeneous sources involves the need to achieve an integration of databases from different data providers within a common platform. Generally, database integration consists of two steps: schema integration and object integration. Concerning schema integration, a crucial part is the identification of semantically corresponding elements in different schemas. This process is referred to as schema matching. In this paper, we present a data-driven approach for the matching of geospatial schemas. It is based on the idea of exploiting the instance-level relations between multiple representations (of one and the same real world object) that have been captured on the basis of different schemas. As a result, correlation measures are derived describing the degree of correspondence between the object classes of different schemas.