This paper presents a statistical scheme for the seasonal forecasting of North China's surface air temperature (NCSAT) during winter. Firstly, a prediction model for an decrease or increase of winter NCSAT is established, whose predictors are available for no later than the previous September, as this is the most favorable month for seasonal forecasting up to two months ahead. The predicted NCSAT is then derived as the sum of the predicted increment of NCSAT and the previous NCSAT. The scheme successfully predicts the interannual and the decadal variability of NCSAT. Additionally, the advantages of the prediction scheme are discussed.