The wavelet network and its improvement version are applied to the temperature compensation of start-up drift of fiber optic gyro (FOG). Based on experiments in the entire temperature range of FOG, influences of inner temperature, temperature gradient and temperature variation rate on start-up drift are analyzed, and the start-up drift model consisting of the above three temperature factors is built. Then compensation based on the model can be carried out to improve the start-up accuracy of FOG. The experimental results indicate that the start-up drift of FOG can be compensated effectively by using wavelet network, and the problem of real-time applications can be settled.