Aiming at the hysteretic nonlinearity of piezoelectric actuator, a modeling method combining least square method with radial neural network is proposed in this paper. Firstly, by setting up the test system of the piezoelectric actuator displacement, the corresponding curve relationship between the output displacement of the actuator and the input voltage is obtained. Then the polynomial fitting of the curve is carried out by least square method to obtain the hysteresis mathematical model of the piezoelectric actuator. On this basis, the radial basis function neural network method is used to optimize the model. Finally, the established model is analyzed. It is found that the maximum error Emax is equal to 0.244 7 μm and the standard deviation δ is equal to 0.059 02 μm by using the polynomial mathematical model fitted by the least square method, while the maximum error Emax is equal to 0.079 89 μm and the standard deviation δ is equal to 0.016 04 μm after optimizing by RBF neural network modeling. The experiment verifies that the model has higher accuracy. This paper provides a new method for modeling of the piezoelectric actuators hysteresis.