Abstract : Currently, vast amounts of geospatial information are o ffered through OGC's services. However this information has limited formal semantics. The most common method to search for a dataset consists in matching keywords to metadata elements. By adding semantics to available descriptions we could use modern inference and reasoning mechanisms currently available in the SemanticWeb. In this paper we present a novel architecture currently in development in which we use state of the art triplestores as the backend of a CSW service. In our approach, each metadata record is considered an instance of a given class in a domain ontology. Our architecture also adds a spatial dataset of features with toponym values. These additions allow us to provide advance searches based on 1) Instance to class matching, 2) Class to class subsuming relationships, 3) Spatial relationships resulting from comparing the bounding box of a metadata record with our toponym spatial dataset.