Developments in remote sensing created new possibilities to capture the human environment on a large scale. This data is used in the generation of large scale Three dimensional (3D) city models, which has led to a higher demand for 3D city models in a wide range of fields. The lack of semantic information in many 3D city models is a considerable limiting factor in their use, as a lot of applications rely on semantic information. This research is a first step in creating an automatic workflow, that semantically labels a plain 3D city model, with level of detail 1 or 2, represented by a triangulated polygon mesh, with semantic and thematic information as defined in the CityGML standard. The first step in this labelling process is the reconstruction of the building entities and (parts of) the terrain: the thematic features. Next, two methods are proposed to semantically label the surfaces in the previously defined building entities. The first implemented method is a best practice of methods that are tested, which aim at labelling the 3D city model with the classes WallSurface, Roof- Surface, and GroundSurface. The second method is an explorative approach and a proposal that additionally recognises the classes OuterCeilingSurface and OuterFloorSurface. In this approach, an proposal is made that extends the current class definitions of a RoofSurface and a WallSurface in CityGML. This research shows that by extending these defintions, a CityGML file with Level of Detail 1 and 2 can be semantically labelled automatically. The results show that a high semantic classification accuracy is possible. The accuracy depends on a number of factors, floating point precision errors is the biggest limitation factor in the thematic and the semantic labelling.