As 3D city models are becoming available and attractive for consumer purposes in personal navigation systems or visualization applications, the hardware used to visualize these models changes from special to consumer PCs or even to handheld and wireless devices. As a consequence, tools have been developed to alter single building models in their LOD in order to reduce the amount of data to be transmitted. While these generalization algorithms construct less detailed representations of the input models, they will also produce inconsistencies. In the case of single stand-alone building models, this may result in mismatches between the original and generalized buildings’ ground plans, causing visual inconsistencies and topological inconsistencies in the interaction with other objects like streets. When adjusting simplified neighboring and adjacent models to the original models’ ground plan, additional errors may occur in features like common façade lines to be lost or the development of unintentional new features like gaps. While these errors visually disturb the transition between different LODs, the topological correctness of the resulting city model may also be degraded by overlaps evolving from the treatment of single buildings throughout the generalization process. Means for preserving visual and topological correctness in the generalization process will be presented in the paper.