Three-dimensional maps are deemed better for navigation purposes as they offer a larger number and more realistic navigation cues than two-dimensional maps. Improvements in two key technologies have opened the doors towards utilization of 3D maps for car navigation devices. Advances in data acquisition technologies and data processing methods have made creating photorealistic three-dimensional city models cheaper and to a large extent automatic, while advances in mobile technologies have made e.g. modern smartphones powerful enough to visualize photorealistic 3D graphics. Despite the latter improvements, making three-dimensional mobile maps remains a challenge due to the large amounts of data and the device’s limited amount of memory and pro- cessing power. These limitations can be overcome by intelligently reducing the amount of information that is handled and displayed by the device. This thesis presents an information reduction and prototyping framework that reduces the amount of information contained in city models so a to enable their loading and display on car navigation devices. The information reduction method consists of two steps. The first step selects buildings that are close to the driver’s route with the idea that these aid the driver in navigating. Buildings that are far from the route are discarded. In the second step, the selected buildings’ external representation is adapted to match their navigational value that is based on their thematic, semantic and cognitive properties. For instance, a building of type ’restaurant’ and ’brand’ McDonald’s offers more navigational cues than a block of gray, anonymous residential buildings. The latter are styled in generic textures whereas the former is styled in photorealistic textures. The relations between a building’s semantic and thematic properties and its external representation are captured in visualisation rules. A prototype is built that implements the designed information reduction methods and tests their effectivity. The selection step is performed using a spatial database while the visualisation rules are processed by an expert system. The reduced 3D scenes are displayed in a game engine that also performs performance measurements. The obtained results are conclusive: the performance of a visualisaion in terms of frame rate and used graphics memory is governed by the the amount of textures, much more so than the number of geometries. Effort should therefore be directed towards the reduction and/or simplification of textures rather than geometries.