By reviewing the existing models for describing gas/particle partitioning of semi-volatile organic compounds in the atmosphere, it was assumed that gas/particle partition coefficient, expressed as
Kp, may be predicted using molecular descriptors. Overall 14 molecular descriptors of each compound calculated using semi-empirical method parametric model 3 (PM3) were tested against log
Kp of selected PCBs to determine the best ones governing partitioning. Eight descriptors molecular weight (
Mw), molecular volume (
Mv), total energy (TE), electronic energy (EE), squared atom electron densities on carbon, hydrogen and chlorine atoms in a given molecule
and average molecular polarizability (αm) were found to be highly correlated with log Kp compared to other molecular descriptors. Using Partial Least-Squares Regression method (PLS), two-, three- and four-descriptor QSPR models with high fitting characters were successfully developed and their robustness and predictive power were further validated by internal cross-validation and external test. Finally, the gas/particle partition coefficients of all 209 PCBs were predicted for the first time.