Temperature is one of the main factors that affect the accuracy of fiber optic gyroscope (FOG), and static temperature characters of FOG are studied through theoretical analysis and experiments. Temperature modeling of zero bias based on typical wavelet network is proposed, which is compared to polynomial fitting, and the accuracy is improved greatly. Further, modification of typical wavelet network is studied, including initialization of parameters and a new momentum adaptive gradient descent method for parameter updating. Through experiments, it is proved that the modified wavelet network can further improve convergence speed and fitting accuracy, which is better in describing the temperature characters of FOG. In order to verify the university of wavelet network algorithms, temperature modeling and accuracy analysis based on testing data of different inertial instruments are carried out, and results indicate that the proposed methods can be used for static temperature modeling of inertial instruments sensitive to temperature.