Abstract: The satellite observation with a resolution of ten meters provides images of earth surface. The precise spectral information allows a classification of earth objects. Due to some considerations, Markov Random Field has become a common search procedure. Attention has been focused on utilizing the spatial context in image classification; labels are to be assigned to individuals pixels or groups of pixels. In this approach, we relied on, among approaches to Markov Random Field to a supervised classification of satellite images. This approach is known as the Multi-Scales model. We use an energy expression according to Potts model.