Abstract: Based on Radial Based Function (RBF) neural network, a method for structured light based 3D vision inspection is presented. The method uses RBF ANN (Artificial Neural Network) to establish the mapping relationship between a real object in the wold and its image captured by CCD camera, i.e., the mapping relationship between frame coodinate and its image coodinate. Compared with common methods, the preset approach ignores the vision model error, and allow the existence of optical adjust error. By overcoming disadvantages of common methods efficiently, higher measuring accuracy can be obtain.