Underwater measurement based on stereo vision attaches great importance to camera calibration. However, it is challenging to perform accurate calibration due to the significant refraction presented at the interfaces of air and water. To solve this problem, a calibration method for an underwater binocular vision system based on the optimized refractive model is proposed. First, conventional calibration is performed to obtain basic initial camera parameters using checkerboard images collected in the air. Then, an evolutionary multi-objective function is established according to Snell's law, the refractive light propagation path, and checkerboard geometric relationship. Finally, precise camera parameters and involved refraction parameters are both obtained for underwater target positioning and size measurement by the non-dominated sorting genetic algorithm of the reference point. A group of experiments is performed, and the validity and effectiveness of the proposed calibration algorithm is demonstrated.