Abstract: A multi-scale fusion algorithm is proposed and applied in GPS/SINS integrated navigation system by combining the traditional analysis method of dynamic system based on model with the analysis method of signal multi-scale transformation based on statistical characteristics. This algorithm first establishes the dynamic equation and observation equation of system in original scale. The dynamic observation equations are then fused on different scales by wavelet transformation. At last the fused results on different scales are fused globally and optimally and are feedback to correct SINS. The experimental results based on practically measured data and theoretical dynamic simulation show that this method can fuse this system efficiently and optimally, and this method can have better precision than the several common filtering methods.