This paper discusses the concentration characteristics of PM 2.5, as well as its relationship with meteorological factors in autumn and winter (from September to the following February), from 2013 to 2018 in the Beijing-Tianjin-Hebei (BTH) region. The accuracy and uncertainty of the air quality forecast models NAQPMS (nested air quality prediction modeling system), CMAQ (community multiscale air quality modeling system), and CAMx (comprehensive air quality model with extensions) were analyzed based on the model-predicted and measured PM 2.5 concentration in autumn and winter from 2015 to 2018. The accuracy of NAQPMS, CMAQ, and CAMx during typical heavy air pollution was also tested. Moreover, methods to improve the accuracy of the model forecast were discussed. The results showed that the mean concentrations of PM 2.5 in the BTH region were 122, 98, 82, 99, and 65 μg· m-3 in the five autumn and winter periods, respectively. When the air quality index (AQI) exceeded 150 during each autumn and winter, it reached 229, 198, 210, 204, and 180 μg· m-3, respectively. There were 64 occurrences of heavy regional PM 2.5 air pollution in autumn and winter from 2013 to 2018. The average duration was longest in the 2013 to 2014 period, and shortest in the 2017 to 2018 period. The peak concentration and average concentration of PM 2.5 decreased year on year, except for the period from 2016 to 2017. In autumn and winter, PM 2.5 concentration had a relatively close relationship with relative humidity, wind and sunshine duration, compared with a weak relationship with temperature and air pressure. Regional heavy air pollution always happened under the condition of low wind speed (less than 2 m· s-1), higher relative humidity (greater than 65%), and southwest and northeast wind direction. In addition, the heavy air pollution of PM 2.5 in BTH in autumn and winter can be effectively forecasted by NAQPMS, CMAQ, and CAMx. The predicted and measured PM 2.5 concentration showed a close relationship. The models performed well in forecasting Zhangjiakou, Chengde, and Qinhuangdao, but by contrast overestimated in Tangshan, Shijiazhuang, Baoding, Beijing, and Tianjin. The uncertainty of emission sources, measured and predicted meteorological data, and the atmospheric chemical reaction mechanism may be the main reasons for the overestimate.