With the fast economic development in China in recent years, air pollutions are becoming increasingly serious. It is, therefore, imperative to develop new technology to solve this issue. Due to the wide spatial coverage of satellite remote sensing, along with the relatively lower cost compared to ground-based in situ aerosol measurements, satellite retrieved aerosol optical depth (AOD) is widely recognized as a good surrogate of surface PM2.5 concentrations. In this study, two years (2007-2008) of AOD data from moderate resolution imaging spectroradiometer (MODIS) onboard Terra at five observational sites of China (Benxi, Zhengzhou, Lushan, Nanning, Guilin), combined with five meteorological factors such as wind speed, wind direction, temperature humidity and planetary boundary height, were used as important input to establish the Back Propagation (BP) neural networks model, which was applied to estimate PM2.5. Afterwards, the model estimated PM2.5 was validated by in situ PM2.5 measurements from the five sites. Specially, scatter analysis showed that the linear correlation coefficient (R) between ground PM2.5 observation and model estimated PM2.5 at Lushan was the highest (R = 0.6), whereas the R values at the four other sites were lower, ranging from 0.43 to 0.49. Time series validations were performed as well, indicating that the R value significantly varied from day to day. However, the R value could be significantly improved by fitting the five-day moving average ground observation values against the model estimated PM2.5 data. Also, the R value at Lushan was the highest (R = 0.83), suggesting that MODIS AOD can be used to monitor PM2.5 by the BP networks model developed in this study.