There exist the spectral information and the spatial information in a remote sensing image. Discriminating a pixel depends not only on its spectral value

but also its spatial position and the relationship between the pixels. The spatial information

existing in an image

is usually subdivided into types-texture and context. This paper is about the classification models using contextual information. Three Markov mesh models are structured

the models then are used for contextual classification. The methods of estimating the spatial correlation parameters are given. A recursive contextual classification procedure

algorithm and programming is presented. The effectiveness and efficiency of the classification models is proved by experiment.