In computer vision, camera calibration as the premise of camera measurement technology, is an essential part. Aiming at the problem that the training accuracy of camera calibration method based on neural network is not high enough, a camera calibration method based on double neural network was proposed. Starting from the imaging model, it was deduced that the camera coordinate ${Z_{\text{c}}}$ was a function of the world coordinate and the pixel coordinate. On the basis of considering ${Z_{\text{c}}}$, the imaging model was simplified into two function relations, and two neural networks were used for calibration, which not only differentiated the task amount of single neural network, but also fully followed the imaging model. The experimental results show that compared with other calibration methods based on neural network, this method improves the accuracy of camera calibration. And the average calibration error is 0.1786 ${\rm{mm}}$ in the calibration range of $400\;{\rm{mm}} \times 300\;{\rm{mm}}$, which verifies the feasibility and effectiveness of proposed method.