The 3D measurement based on stereo vision, as a new measurement technology, has been widely used in reverse engineering, robot vision and quality control, etc. In this paper, the basic principle and method of calibrating 3D measurement system based on stereo vision respectively using genetic algorithm and neural network was introduced. Then the measurement system was calibrated and the precision of 3D measurement system was compared. The experimental results indicate that the method based on BP neural network can improve the precision of measurement better.