Abstract: To improve the accuracy of a strapdown north finder used under dynamic conditions of disturbances, a new wavelet denoising method is introduced to process the original output of the north finder. From wavelet transform modulus across scales, it is seen that useful signal is stable and some maximum modules does not decrease in larger scales. So maximum modules and soft-thresholding are proposed.Some maximum modules whose values grow larger with increase in their scales are substituted by mean value of non-maximum modules adjacent to them. Thus disturbances whose Lipschitz exponent is larger than zero are filtered. Soft-thresholding is used to filter the noise whose Lipschitz exponent is smaller than zero. The method can improve the accuracy of strapdown finder to 1 mrad.