Abstract: In this study,the concentrations of PM10,CO,NO2,O3,O3(8h) and SO2 concentrations have been employed to predict the concentration of PM2.5 using multivariate statistical and time series analysis methods.The data have been collected from seven monitoring stations in Urumqi(Xinjiang province,China) from November 2014 to March 2015 (winter).The bivariate correlation analysis shows that the outdoor PM2.5 concentration is highly correlated with that of PM10,CO,NO2 and SO2.Further,using exponential smoothing model to predict the stability of PM2.5 concentration, the best smoothing index is 0.32.Principal component regression (PCR) has been used for modeling a prediction the instability of PM2.5 concentration and values is 0.803 for explaining the instability of PM2.5 concentration.The performance indicators of models indicate that the prediction for PM2.5 has an effective prediction.