In the text，a robust undecimated wavelet domain noise filtering method for magnetic resonance （MR）images was proposed.Ａ single parameter was used to balance the degree of noise reduction and preservation of relevant details.The algorithm developed a generally valid knowledge about significant image features across the wavelet scales to perform a preliminary coefficient classification，which was used to empirically estimate the statistical distribution of image features or noise coefficients.A wavelet domain indicator of the local spatial activity achieved the adaption to the spatial context in the image.The results show its efficiency in MRI denoising.