Real-time and web-based visualisation systems like digital city or earth viewers have gained significant public interest in recent years. Due to advances in software, hardware and network bandwith, they are nowadays capable of streaming and presenting spatial information at near photorealistic quality. Such accurate illustrations that are true to detail are, however, not always the most adequate tool to communicate spatial information. Particular location based services and context-aware applications that usually run on mobile devices like personal digital assistants or mobile phones require special presentation techniques to make spatial situations easier to perceive and comprehend. Similar intentions are pursued in the creation of cartographic and map-like presentations where specific requirements about the minimum object and feature size must be met. If these are not satisfied, the objects or even the whole scene must be transformed with the help of generalisation operations. One important operation for the generalisation of three-dimensional urban scenes is the simplification of building models. Many simplification algorithms for general surfaces are already known from the field of computer graphics. For cartographic presentations, however, they are not applicable as they are unable to guarantee the preservation of object properties during their transformation that are specific to buildings. These are the parallel and right-angled arrangements of façade walls and the symmetries of roof structures. Within this paper we put our focus on the simplification of three-dimensional building models and present new generalisation algorithms that are specific to this kind of data.