Abstract: Friction is the main nonlinear disturbance that affects the low-speed operation and point-to-point positioning accuracy of the linear servo system. Although the friction model of Generalized Maxwell-slip (GMS) can accurately describe the friction characteristics for feed-forward compensation, it has the problem of oscillation during the transition of the switching point, and it is susceptible to measurement noise and changes in friction parameter. In this paper, a friction compensation strategy based on the combination of smooth GMS model and disturbance observation method is proposed. First, the hyperbolic tangent function is used to replace the switch function to solve the problem of repeated crossing during the transition of the switching point in the GMS model, and an offline identification method of the model is proposed. The smooth GMS model is continuous and infinitely differentiable in the whole defined domain, which is more suitable for introducing state observers. Second, aiming at the accuracy problem of offline friction compensation, a fourth-order linear extended state observer based on model information is designed to compensate for residual friction and unknown disturbance, and the Chebyshev filter that is close to the ideal filter characteristic is selected to design the observer gain to solve the problem of low gain in passband. In order to verify the effectiveness of the proposed strategy, a positioning experiment is carried out on a small high-precision permanent magnet synchronous linear motor positioning platform. The experimental results show that the proposed method can effectively suppress the friction during speed switching, and greatly improve the tracking accuracy.