This study estimates infiltration factors (F-inf) and ambient personal exposure factors (F-pex) for fine particulate matter (PM2.5) in two Chinese megacities, and constructs predictive models to explore their determinants. Personal-indoor-outdoor PM2.5 filter samples were collected for five consecutive days from 33 residences (of retired adults) in Nanjing (NJ) and Beijing (BJ), China, in both the non-heating season (NHS) and the heating season (HS). Elemental sulfur in filter deposits was determined by energy-dispersive X-ray fluorescence for PM2.5 F-inf and F-pex estimations. Season-specific models developed by stepwise multiple linear regression were evaluated using R-2 and root mean square error (RMSE). The median [interquartile range (IQR)] of F-inf varied from 0.76 (0.15) in the HS to 0.93 (0.11) in the NHS in NJ; and from 0.67 (0.16) to 0.86 (0.12) in BJ. Similarly, F-pex was significantly higher during the NHS [NJ: 0.95 (0.07); BJ: 0.89 (0.14)] than during the HS [NJ: 0.76 (0.17); BJ: 0.67 (0.11); p <0.0001]. Common predictors of F-inf and F-pex included window opening behaviors, meteorological variables, and building age. Moreover, air conditioning and distance to the nearest major road had an influence on F-inf, while predictors of F(pex )were more related to human behavior and activity (e.g., time spent outdoors and transportation). The models accounted for 35.4%-68.1% (RMSE: 0.065-0.101) and 41.6%-77.0% (RMSE: 0.033-0.103) of the variance in F-inf and F-pex respectively. By indicating the determinants of F-inf and F-pex these models can improve ambient PM2.5 exposure assessment and reduce exposure misclassification.
