Abstract: A technique of speckle reduction in synthetic aperture radar (SAR) images was presented. Based on the character of wavelet transforms that the approximate coefficients of the transformed image still obey Gibbs distribution, the simulated annealing to the wavelet domain was introduced. The traditional logarithmic transformation was modified to adapt the power SAR images, which can maintain the mean of the SAR image better and achieve a better equivalent-look, then based on multi-resolution analysis theory and wavelet transform theory, an optimal mother wavelet was found to match the specific images. The algorithm of simulated annealing was applied to the coefficient in the scaling subspace and the algorithm of soft-thresholding was applied to the coefficient in the wavelet subspace distinctively. The result of experiments show that the new technique achieves a better smooth result when compared to the traditional wavelet denoising, at the same time it saves much time when compared to the simulated annealing, and it improves the structure and edge preservation with satisfying denoising results.