In order to detect the actrul changes from two remotely sensed multispectral images gained at the same location and at different times, a statistical method is created —Statistical Test. In the method, local regression is applied to the areas of interest from the two images which are gained from two times. And when the difference of regression coeffients are very large, these areas of interest are subdivided into more smaller areas. Compared with the conventional methods, the proposed method is not only undependent of the impulse noise, but also the method is not influenced by the apparent changes caused by the sensitivity degradion in sensors or the location changes of the light source.