Model registration is widely used in localization and inspection of parts. It remains a challenging problem in some situations. The density of the measured points of the parts depended on the surface curvature in 3D optical scan measurement technology. The local surface in small curvature was to enable more accurate representation with comparatively sparse measured points. Thus the area authorized by each point was different. So each point should be endowed with different status in registration progress. In addition, for a complex part with multiple free form surfaces, different precision requirements were often specified to different regional surfaces in design, and manufacturing process might make the precision difference in regions. Without consideration of the difference, the existing registration methods were prone to obtain the result which balanced the surface error. In the case, the error of the surface with high precision became larger than the physical truth, and the error of low precision regions turned out to be just the opposite. Based on the two aspects, a registration method based on composite weighting parameters of measured point area and regional geometric accuracy was proposed. Composite weighting parameters were constituted by estimating regional precision iteratively and calculating the area of the Voronoi diagram of each measured points. They controlled the influence of different data on the registration. As a result, the registration conformed more to the real engineering. Both theoretical and experimental results verified the efficiency and availability.