Based on the concept of data reuse and data sharing, a 3D building model retrieval approach is proposed to reconstruct point clouds for cyber city modeling and updating. Thanks to age of Web 2.0, an increasing number of models are available on the website like Google Warehouse. A huge database with a great diversity can be easily constructed from the open sources of these platforms. We aim to build a 3D building model search engine for the demand of following application such as quick modeling instead of those with large time-consuming. Spherical harmonics function is chosen as the shape-descriptor to parameterize the models in low frequency domain. The most similar model can be extracted by matching the parameterized spherical harmonic coefficients between models and input data. Point cloud data obtained by airborne LiDAR is inputted as query to search the similar models from database. Properties of point cloud data with incompleteness and noise is also a challenge to this research. A set of data preprocessing procedures will be executed to optimize the retrieval result. The experiment results show the possibility and flexibility of proposed algorithm to effectively retrieve the fittest model from database.