Synthetic aperture radar (SAR) image is likely to be disturbed by a speckle noise due to the radarwave coherence. According to the particularity of speckle noise in SAR image, logarithmically transformation was first applied, and àtrous algorithm was used to decompose the image. Then the manipulation of the wavelet coefficients was performed based on Bayesian estimator. The result is compared with adaptive filtering algorithm based on local statistics and filtering algorithm based on Mallat decomposition. It is shown that the approach has the advantage both in speckle suppression and in preservation of edge information.