A cyber city is an effective way to visualize the reality in a cyber space. 3-D building models are one of the core elements in a cyber-city. As Lidar point clouds are usually employed to reconstruct and evaluate the building models, the registration of 3-D building models and lidar point clouds is an essential work to ensure both data are in a unify system. In this study, we perform the data co-registration of 3-D building models and lidar point clouds using Least Squares 3-D (LS3D) Surface Matching algorithm. This method iteratively minimizes the 3-D Euclidean distance between these two surfaces using Least Squares Adjustment. First, we find the initial conjugate surface between these two data. Then, a seven parameters 3-D similarity transformation is established to compensate the registration error. All the calculated differences are applied to obtain the transformation parameters using Least Squares Adjustment. Finally, all the lidar points are transformed to the building models space. The test area is located in Taipei, Taiwan. The input data are 3-D building models generated from stereo aerial images and the airborne lidar acquired by Leica ALS 50 with 10 points per meter squares. As the accuracy of lidar data is higher than the building models, the shaping errors and the transformation parameters from Least Squares 3-D Surface Matching can be treated as a quality index of building models. Moreover, the area with large shaping error may represent the missing part of the building. The registration of 3-D building models and lidar point clouds is beneficial to accuracy analysis and model refinement of building models.