Abstract: A new classifying algorithms,BAyes Normal Dynamic Clustering,abbreviated as BANDC,is presented in this paper,chiefly for geographical environments recognition from remot sensing images.It has two advantages:1)the ability to estimate the parameters of the classifier from the images by self-discipline;2)as a special Bayes classifier,it can estimate prior probabilities of W_i by the computer and pick out ill-point,thus increasing the robustness of the classifier.The knack of this algorithms is the estimation of the parameters from the grey level histogram by using fitting techniques.It involved the Bayes rule criterion function and the estimation of the peak positions of the histogram.