This paper presents the method for the generation of clouds based on GMS 5 images.A statistical relation between multispectral threshold and temperature on isobaric surface is obtained in the line of the optimal regressive statistics,and a dynamical changeable multispectral threshold is calculated by the virtue of the forecast of MM5 meso scale NWP model in order that the discriminant threshold of cloud adapts to the changeable weather systems.The experiment shows that the method of multispectral threshold is a feasible technique for cloud differentitation and classification,which can exactly distinguish cloud and clear sky,high and low clouds as well as semitransparent or fractional cloud and middle cloud .This technique can be used in the operational cloud classification.