In this paper, based on the “Mesoscale Numerical Forecasting System” developed by the Linyi Meteorological Bureau, the BP neural network model was applied to forecast the surface air temperature using the high resolution numerical forecasting data and observation as input. The result shows that accuracy rate of the BP model is higher than the stepwise regression model and MM5 model in surface air temperature forecasting, and the BP model can be applied in realtime surface air temperature forecast