In order to adapt spatial nonstationary character of an image, a denoising method based on Wavelet-Domain Classified Hidden Markov Tree Model(CHMT) is proposed.In this method,image's coefficients of every scale and subband are divided into C groups to emphasize the spatial nonstationary character,so that one image corresponds with C HMTs.Then these coefficients are initialized,trained by EM algorithm and inverse-transformed.Test result shows that this method improves image quality(PSNR) obviously while calculation doesn′t add.