The community multiscale air quality (CMAQ) model was used to forecast air quality over the Pearl River Delta region from December 2013 to January 2014. The pollution forecasting performance of CMAQ coupled with two different meteorological models, i.e., the global/regional assimilation and prediction system (GRAPES) and the fifth-generation mesoscale model (MM5), was assessed by comparison with observational data. The effects of meteorological factors and physicochemical processes on the forecast results were discussed through process analysis. The results showed that both models exhibited good performance but that of GRAPES-CMAQ was better. GRAPES was superior in predicting the overall variation tendencies of meteorological fields, but it showed large deviations in atmospheric pressure and wind speed. This contributed to the higher correlation coefficients of the pollutants with GRAPES-CMAQ but with greater deviations. The underestimations of nitrate and ammonium salt contributed to the underestimations of both particulate matter and extinction coefficients. Source emissions made the only positive contributions to surface layer SO₂, CO, and NO. It was found that O₃ originated primarily from horizontal and vertical transport and that its consumption was predominantly via chemical processes. Conversely, NO₂ was found derived primarily from chemical production.