Based on the output data of model RegCM3 during the heating period of 2005 and 2006, and the daily monitoring data of Zhengzhou Municipal Environmental Monitoring Center, a prognostic equation is constructed to forecast the mass concentrations of air pollutants PM₁₀, SO₂ and NO₂ by using the stepwise regression method. But the application of the method to forecasting tests of the pollutants' mass concentrations in the heating period of 2007 is not satisfactory, and the forecasting accuracy is significantly lower than the historical fitting rate. In order to improve the forecasting accuracy, according to deficiency of the statistical method(the correlations among forecasting factors are not considered when selecting them, so the nonorthogonalities among them result in regression instability and more errors), a new forecasting model is proposed by using the empirical orthogonal function(EOF) combined with the multiple linear regression analysis, and the daily mean mass concentrations of pollutants are chosen as the forecasting object during the heating period. Results show that the new method can improve the forecasting accuracy of air quality in a certain degree.