A method on boundary feature extraction of complex surface fitting was proposed due to the fitting optimization during the assembly of auto-body closure parts. The extraction objects, sampling size and searching neighbor were decided based on the fitting boundary features. Point cloud was sampled and constructed by k-d tree. The points within the searching neighbors were projected to their least square tangent plane and the boundary points with fitting features could be extracted by judging the distribution. A case study was conducted to prove the effectiveness and efficiency of this method.