Abstract: A new method for change detection based on multi-scale fusion was proposed in this paper to make full use of scale information of SAR images and improve robustness and steadiness of change detection algorithms simultaneously. The proposed method started with wavelet decomposing of log-ratio difference image, and then obtained reconstructed images under different scales independently. After that, the mean-iterative threshold selection was adopted to separate the changed pixels versus the unchanged ones. At last, a fusion means based on Markov random field, was introduced to achieve the final change map that integrates results of different scales. The means taken in this paper not only effectively take full advantage of scale information but also detect more details. Experiments on two pairs of real SAR data confirm that the proposed method is able to improve accuracy and robustness of change detection effectively.