Abstract: Change detection in muhitemporal remote sensing images is a challenge because of the illumination variation and noise. In this paper, we develop a robust change detection algorithm based on muhiscale analysis and multivariate hypothesis test. A modified wavelet transformation is used to decompose the difference image of two co-registered remote sensing images into four sub-images. The modified wavelet transformation has no down-sample procedure, and the sub-images have the same size and coordinate system with the input images. Then the four sub-images are integrated into a vector matrix, with each vector is composed of four weighted values that are at the same space position of four sub-images. The two vector matrix can be analysised under the frame of multivariate hypothesis test. Experiment results are carried out to prove its superiority.