Abstract: In this paper, four types of predictors were proposed from the actual situanon of relationship between predictand and predictors, then the optimal predictors were selected using the principles of optimization method, at the same time the optimal prediction equation was selected based on the method which was used to the forcing introduction of some important predictors. The computing results from two examples show that the optimal predictors were mainly nonlinear, so that the prediction equations builded from stepwise regression or step-wise discrimination method were also nonlinear.The authors bear out that the prediction acuracy of the nonlinear equation is better than the linear one.Finally,the authors also indieate this nonliear prediction techniques might extend to various aspects relating to the quantitative and qualitative forecasting works on multivariate analysis.