In this paper, a method to select some nonlinear predictors and to build a nonlinear forecasting equation was proposed based on optimization principles. The experiments show both the nonlinear predictors and corresponding equation based on this method are better than those linear equations. Finally, the authors indicate that this method can be extended to building of nonlinear discriminatory equation when predi-ctands are discrete. This method is applicable to various forecastings of long-, medium- and short-ranges.