Based on the concept of data reuse and data sharing, a 3D model retrieval and assessment approach is proposed to reconstruct point clouds for cyber city modeling and updating. The main idea is to build a gigantic database containing great diversity of 3D building models. The models in database are collected from model-sharing www applications. All the models in the database are encoded by a small set of low-frequency spherical harmonic functions (SHFs). A point cloud obtained by airborne LiDAR is inputted as query to search the similar models from the database. By means of matching the SHFs coefficients between point clouds and 3D models, the most similar model can be efficiently extracted. The extracted model can be used as a template model to fit the point cloud. The experiment results show that the proposed approach can efficiently extract the fittest model from a huge database. This makes the proposed approach feasible to efficiently construct and update 3D city models.