The classified statistic model of urban ambient air quality forecasting was established by means of pollutant concentration data from June 2000 to December 2002 in Shanghai, and its forecast capability was tested by 1-year (2003) data. Variance analysis was selected to classify the samples, empirically taking season, surface wind direction and precipitation as 3 indicators: winter-half-year and summer-half-year sub-samples (season), west wind and non-west wind sub-samples (wind direction), and rainfall and non-rainfall sub-samples (precipitation). After the 3-layer classification, the whole sample was divided into 18 sub-samples. The high and low concentration sections of the sample classification are practically in accordance with the season, surface wind direction and precipitation distributions of pollutant concentrations. Finally the classified sub-samples were processed to establish a set of forecasting models by the linear successive regression technique. In the capability testing, the classified statistic model has proved feasible in the urban ambient air quality forecasting with higher prediction accuracy compared to the total sample statistic model.