Recent years, BP model has been applied to atmospheric pollution forecast, a neural network prediction model of atmospheric pollutant concentration is set up. The research results show: the absolute errors of TSP between calculating and the monitoring is from 4×10^-3mg·m^-3to 3×10^-2mg·m^-3, the absolute errors of NO_X between the calculating and the monitoring is from 5×10^-3mg·m ^-3 to 2×10^-2mg·m^-3. The correlation between results of calculating and the monitoring is very well. As one of the neural network models, BP model has been applied widely, which is a simple and effective algorithm.So, BP neural network model has supplied a new way for the air pollution forecast.