Aimed at the difficulty in forecasting the positioning accuracy of machine tool feed system,a new model has been established, analyzing the positioning accuracy degradation of the mechanical transmission system of feed system. Firstly, a computer numerical control(CNC) machine tool feed system model is made using a dynamic simulation software Adams to obtain positioning accuracy values for different initial state. Then, a mapping model is set up between the positioning accuracy and the gap, the ball tilt and the workpiece loading by BP neural network. According to the mapping model, the prediction method of the positioning accuracy is discussed in detail. Finaly, the method is proved to be correct and effective by conducting the test experiment of positioning accuracy on the test platform of precision motion reliability.