In this paper we present an approach to introduce semantics into a SLAM-generated 3D point cloud map from 3D laser scans of an office environment. For semantic classification we propose to first reconstruct surface planes in the point cloud. Using an OWL-DL ontology, we automatically analyze relations between surface planes in a cluster to generate hypotheses for present objects. To check these hypotheses we surface sample appropriate CAD models and use standard ICP to fit the scan data with the model of the hypothesized object. The final result is a hybrid semantic map, in which all identified objects have been replaced by their corresponding CAD models.