We applied a global 3-D chemical transport model (GEOS-Chem) to examine the variations in the frequency and intensity in severe winter haze days (SWHDs) in Beijing-Tianjin-Hebei (BTH) from 1985 to 2017 and quantified the roles of changes in anthropogenic emissions and/or meteorological parameters. Observed SWHDs were defined as the days with daily mean PM.sub.2.5 concentration exceeding 150 Âµg m.sup.-3, and simulated SWHDs were identified by using the same threshold but with adjustment on the basis of simulation biases. Comparisons between the simulated SWHDs and those obtained from the observed PM.sub.2.5 concentrations and atmospheric visibility showed that the model can capture the spatial and temporal variations in SWHDs in China; the correlation coefficient between the simulated and observed SWHDs is 0.98 at 161 grids in China. From 1985 to 2017, with changes in both anthropogenic emissions and meteorological parameters, the simulated frequency (total severe haze days in winter) and intensity (PM.sub.2.5 concentration averaged over severe haze days in winter) of SWHDs in BTH showed increasing trends of 4.5 d per decade and 13.5 Âµg m.sup.-3 per decade, respectively. The simulated frequency exhibited fluctuations from 1985 to 2017, with a sudden decrease from 1992 to 2001 (29 to 10 d) and a rapid growth from 2003 to 2012 (16 to 47 d). The sensitivity simulations indicated that variations in meteorological parameters played a dominant role during 1992-2001, while variations in both emissions and meteorological parameters were important for the simulated frequency trend during 2003-2012 (simulated trends were 27.3 and 12.5 d per decade owing to changes in emissions alone and changes in meteorology alone, respectively). The simulated intensity showed a steady increase from 1985 to 2017, which was driven by changes in both emissions and meteorology. Process analysis on all SWHDs during 1985-2017 indicated that transport was the most important process for the formation of SWHDs in BTH with a relative contribution of 65.3 %, followed by chemistry (17.6 %), cloud processes (-7.5 %), dry deposition (-6.4 %), and planetary boundary layer (PBL) mixing (3.2 %). Further examination showed that SWHDs exhibited large interannual variations in frequency and intensity, which were mainly driven by changes in meteorology. The results of this study have important implications for the control of SWHDs in BTH.