For 3D dimensional inspection systems, point clouds measured on each viewpoint need to be processed for quality evaluation. Three steps are usually included in this process: filtering, registration, and error map generation. For quality control, small defects like dints and dents have to be kept in the point cloud. Therefore, a filtering algorithm is required to automatically remove outliers and keep dints/dents. Many filtering algorithm smooth the point cloud for better display, however, since the measured point cloud is used to represent the shape of the part, modification of any point's coordinates is not allowed because that will modify the error map. A point cloud filtering algorithm is developed using a link clustering algorithm to identify and remove outliers. Point cloud filtering is especially important in an iterative closest point (ICP)-based robot hand- eye calibration method because outliers will bring calibration errors into the calculated transformation matrix. With this technique, the cleaned point clouds can be directly transformed to a world frame for registration. This registration method has two advantages compared to feature-based registration methods: 1) the entire inspection process can be automatically executed, 2) avoid holes in point clouds caused by artificial markers. For error map generation, a point-to-plane distance is used in this paper which calculates the distance of a point to its closest triangle. The introduced automated inspection system had been implemented on a PUMA robot system. Experimental results are described in this paper.