The paper analyses the potential of 3D CityGML geometry models for building heating demand simulation. Various aspects of processing 3D data to make it suitable for building simulation are discussed. To validate and calibrate the simulation models, block level aggregated heating consumption data for the city district of Stuttgart Stöckach was used in conjunction with other data such as weather data. The work tries to identify sources of problems in the workflow of using 3D city models and other data bases such as building physics or usage libraries which can lead to a deviation between simulated and measured demand. Depending on the type of CityGML data, adjustments might be necessary to account for missing terrain intersection curves, for wrong wall or roof attributes etc. Only if the 3D model is of sufficient quality, heating demand simulations can yield realistic values. For the case study district the corrections improved the initial mean error of 61% to 29%. A higher accuracy could be achieved for building blocks which have a high share of residential buildings.