The method of correspondence analysis is similar to factor analysis and ma incomposition analysis,which do not contain response variable,and are most app licable to quali tative data or frequency data. It is veryimportant for this type method to selec t variables and samples. In view of the unequilibrium of var iates between model units and prediction units in mineral resources prediction, this paper puts forward the method of stepwise correspondence analysis, which has sol ved the problem caused by choosing variates and samples in the statistical model wi th an irresponding variates, and discusses the optimum allocation of sam ples and variates