||Increasing quality requirements urge the car manufacturing industry to continuously improve their product quality in all possible aspects. The sheet metal of car bodies will have small dents and ripples not able to be seen until varnishing, which will seriously undermine the quality of the product. Detecting and repairing this kind of defects usually cost a lot of labor work   . We developed a new automatic dent detection system which can automatically obtain the surface information of the car bodies, extract features, reduce noise effects and distinguish dents from features and noise. The system will also generate position information for later marking robots. In our system, we use 3D laser scanner as the data acquisition device, and stream those obtained range images into our dent detection kernel. In our dent detection kernel, the normal distance method is used to locate dent candidates, and the geodesic distance method is used to compensate the disadvantages of the normal distance component. We also introduce several denoise filters such as Multi-resolution Bezier Patch filter for data pre-processing, the automatic threshold filter. A feature categorizer is designed and used for distinguish dents from feature lines and noise residues. In the feature categorizer, the Spanning Tree Searching Method is used to cluster points that are spatial connected, and the Seed Growing Search Method is developed to recognize and reconstruct the feature lines from broken line segments caused by the jagged point problem due to resolution issues. Finally, the positions of dents are calculated and output to the marking robot.