For real-time and on-line monitoring DOAS (differential optical absorption spectroscopy) system, a model based on an improved Elman network for monitoring pollutant concentrations was proposed. In order to reduce the systematical complexity, the forecasting factors have been obtained based on the step-wise regression method. The forecasting factors were current concentrations, temperature and relative humidity, and wind speed and wind direction. The dynamic back propagation (BP) algorithm was used for creating training set. The experiment results show that the predicted value follows the real well. So the modified Elman network can meet the demand of DOAS system’s real time forecasting.