In order to compensate for the errors of Fiber Optic Gyroscope (FOG) in an attitude measurement system, according to dynamic and static error distributions of FOG in the time and frequency domains, the basic methods of wavelet, analyses are introduced and how to select threshold and threshold function is compared. The algorithm for real-time wavelet threshold filtering with moving window is adopted to de-noise output signals of FOG. Experimental results show the feasibility of the real time wavelet threshold method in FOG signals de-noising. The real-time wavelet de-noising algorithm with combination of soft and hard thresholds has the advantages over single algorithm, the algorithm is able to not only improve the precision and stability of the FOG, but also provide reference value for dealing with the FOG signals.