A new synthetic aperture radar (SAR) image filter method is proposed based on hidden Markov model (HMM). First the hidden Markov tree(HMT) model in wavelet-domain is established. The parameters can be estimated by EM algorithm.Then SAR image's statistical property is studied, the multiple speckle noise of SAR image can be approximate as white Gauss addition noise under the local area. The wavelet coefficient of signal can be estimated by minimum mean-square estimate(MMSE). Applying this method to simulated image and real SAR image, the result shows that this method can effectively reduce speckle noise, while keeping image details.