The daily average concentrations of PM10，SO2，NO2 and meteorological factors (precipitation and wind) of the corresponding time from 2001 to 2007 in Xi’an, Shaanxi Province, are analyzed with Empirical Mode Decomposition (EMD). The periods are tested and the variance contribution rates are calculated of each intrinsic mode function (IMF) by the maximum entropy spectral analysis as well, for improving the accuracy rate of the urban air quality forecast system (CAPPS2) and instituting the correcting rules for the numerical forecast model. The results show: the main period of PM10 concentration fluctuation is about a week, has clear weekendeffect, and matches well with major scavenging process, but the concentration fluctuations of SO2 and NO2 do not synchronize the scavenging processes, suggesting that the pollution of Xi’an is emissionguided, and the correction rules based on the weekendeffect are ideal for the improvement of PM10 and SO2 numerical forecast results, and uncertain for NO2.