In this paper, we present a technique that generalizes 3D building groups of virtual 3D city models according to a cell structure that is derived from infrastructure networks. In addition, the technique handles vegetation areas and outliers such as landmark buildings. zed 3D representations abstract from complex, detailed 3D city models and enable storage, analysis, exploration, and interaction at varying levels of scales. Our technique implements the cartographic generalization operators clustering, aggregation, and accentuation; it performs the generalization in four steps: 1) City model components are clustered based on the cell structure. 2) For each cell, the weighted average height is calculated, which is also used to automatically identify outliers. 3) Free space is subtracted from the cells such as in the case of outliers or vegetation areas. 4) The modified cells are extruded to building blocks; vegetation areas and outliers are modeled or, respectively, integrated separately...