The practical exploitation of SDI (Spatial Data Infrastructures) raises number of issues as far as it grows. Among them is the heterogeneity of data sources and thus the difficulty for GIS users not to depend on the data source format and of course to learn different systems. This a major flaw with respect to reuse and data sharing. The purpose of our work is to propose a new semantic layer derived from the SQL language that is independent of the underlying data source. This layer, called GDMS (Generic Data source Management System) can first be seen as an abstraction layer between data sources and the SDI tools. We will also show how this layer extends both SQL and spatial semantics and improves the exploitation of the SDI, by providing feedback both in terms of work and data reuse. A simple example mixing heterogeneous data sources will be presented.