In order to better study the machine tool stiffness,a modeling method was proposed to predict the stiffness of mechanical transmission system by measuring the machining error of S-test part,and the stiffness model was established by residual fully connected neural network.The contour error of S-test part was analyzed by using multi-body dynamics,and the error model of transmission system was built.The contour error of S-test was trained by using residual fully connected neural network to get the stiffness of transmission system.The feasibility of the neural network was verified by building the stiffness measurement test-bed and measuring the stiffness of machine tool.The results show that the new modeling method is precise in predicting the stiffness of mechanical transmission system.The residual fully connected model converges faster,with a prediction accuracy of around 80% after 70 iterations.