The output accuracy of fiber optic gyroscope (FOG) in the strapdown inertial measuring unit (SIMU) is significantly influenced by temperature, and the temperature drift compensation must be carried out in practical applications. The bias drift in FOG has shown strong nonlinear characteristics in terms of the change of the temperature, the traditional multivariate linear regression method is hard to meet the requirement of compensation accuracy. Therefore, the wavelet neural network is applied here to establish the temperature compensation model of FOG. By simulating the static drift data of FOG for certain model of SIMU, the experimental results show that the compensation effect based on the wavelet neural network model is better than the traditional multivariate model, and the precision of the FOG is improved effectively.