A significant decrease in surface PM2.5 concentrations has been reported since the implementation of the Air Pollution Prevention and Control Action Plan in 2013. In this study, we use the GEOS-Chem model to simulate the trend in surface PM2.5 pollution in China from 2013 to 2017, as well as the relative contributions of emission reduction and meteorology. The simulated decline rate averaged over monitoring sites in China is around −4.7 μg m−3 yr−1 in comparison with the value of −6.4 μg m−3 yr−1 from observations. The model also captures the variations over different regions, with r in the range of 0.85–0.95. Based on the sensitivity tests against emissions and meteorology, the study finds that the decline in PM2.5 concentrations is mainly driven by the reduction in anthropogenic emissions. The variation in open biomass burning (OBB) is not significant, except in Northeast China (NEC) and Pearl River Delta (PRD), where the changes originated from OBB are 40% and 30% of those associated with anthropogenic emission reductions. Changes in meteorology from 2013 to 2017 led to significant increases in PM2.5 concentrations in most areas in China, except in NEC. The increase attributed to meteorology, to a large extent, could be explained by the significant decrease in surface wind speed (WS) and planetary boundary layer height (PBLH) between 2013 and 2017, combined with their negative correlation with PM2.5. The decrease in PM2.5 concentrations in NEC, on the other hand, could be explained by the significant decrease in relative humidity (RH) there combined with the positive correlation of RH with PM2.5, while the changes in WS and PBLH there are relatively small compared with other areas. The change in meteorology, therefore, hinders the improvement of air quality via emission controls in most of China. In Sichuan Basin (SCB), the increase due to meteorology almost compensates for the decrease associated with emission reduction, leading to the least change in PM2.5 concentrations, although the decrease due to emission controls is the largest compared with other areas.