An adaptive threshold estimation method for multiplicative speckle denoising in SAR images was proposed, based on wavelet domain merging with translation invariant denoising. The subband threshold was derived using bayesian estimation theory and a new probability density function. The method was applied for speckle SAR images by using logarithmic transformation. Theoretical analysis and experimental results demonstrate that the method is computationally more efficient, and displays better noise removal as well as better preservation of sharp features.