Abstract. Three dimensional (3D) city models have many uses including city walk-throughs or fly-throughs to show what a new building would look like in situ, or whether a view or light will be blocked by a new structure, flood and signal modelling. Often, these models are created using a process of extrusion of detailed 2D topographic mapping. The resulting 3D datasets contain many thousands of polyhedra, which in turn results in performance issues when attempting to visualize such models in Google Earth. To address this issue we first generalise the 2D dataset (using simplification and aggregation) and then extrude the generalised 2D maps to 3D, comparing the rendering performance and visual aspects of the resulting datasets.