Abstract: Based on the forecasting products of CUACE (CMA Unified Atmospheric Chemistry Environment) and CMAQ (Community Multiscale Air Quality) models,integration forecast models for PM_2.5 at different positions in Shenyang under conditions of small wind speed and high relative humidity were developed and validated using an artificial neural network method with back-propagation (BP) algorithm.The results indicate that PM_2.5 concentrations predicted by integration models are much closer to their observational values than those predicted by CUACE and CMAQ.The values of mean deviation and NMSE (Normalized Mean Square Error) of modelling results decrease significantly,and the values of FAC2 increase obviously.The PM_2.5 forecast from integration models can better reflect the variation of high PM_2.5 concentrations,and its development at surrounding urban areas of Shenyang is significant.The integration models based on BP artificial neural network are a kind of effective method for PM_2.5 forecast,which can provide a reference to the real-time operational forecast of air quality.