In this paper, we apply the method of Granger causality, which is more accurate than classical correlation analysis method, to determine whether the main air pollutants--Nitrogen oxides (NO(x)), SO(2) (Sulfur Dioxide), CO (carbon monoxide), TSP (total suspended particulates), PM(10) (particulate matter smaller than 10 microns)--and the mortality of respiratory diseases of the residents in Beijing have causal relationship. After ensuring NO(x), SO(2) and CO as the responsible substances, we use the time series method to construct the autoregressive integrated moving average model (ARIMA) of the pollutants, so that we could predict the amount of the pollutants from 2005 to 2008. Then we use the predicted value of pollutants as the input of the neural network model and obtain the output as the change of the death rate of respiratory diseases from 2005 to 2008. In the end, reducing the amount of pollutants by 10% and inputting the data in the neural network model, we make the prediction to evaluate the level of the pollutants and concluded that NO(x) is the most important pollutant to control.