The mathematical model of a binocular vision sensor is analyzed. A novel calibration method for binocular vision sensor based on matching the synthetic images of concentric circles is proposed. The target of concentric circles is randomly placed in the field of vision sensor for several times and the images of target are acquired by two cameras. The synthetic images are calculated through the model of camera and the known position of concentric circles in the target plane. By matching the synthetic images and observed images, the coordinate of the centre circles in image plane is located with optimization. Then, based on the image coordinate of the concentric circles in two cameras and the constraint of binocular vision, the parameters of binocular vision sensor are optimized by Levenberg-Marquart method. Based on the calibration proposed by Z.Y.Zhang, the whole information of the images is optimized. The experiment shows that the accuracy of calibration is improved.