Optimized calculation of 209 PCBs was carried out at B3LYP/6-31G* level in gaussian98 program. Based on the theoretical linear solvation energy relationship (TLSER) model, the obtained structural parameters were taken as theoretical descriptors to establish the novel QSPR model for predicting n-octanol/water partition coefficients (lgK ow ) of PCBs. The new model achieved in this work contains three variables, polarizability α, E LUMO and E HOMO , of which r 2 =0.9484, SD=0.18, with larger t values. In addition, the variation inflation factors (VIF) of variables in the present model are all less than 5.0, suggesting high accuracy of the lgK ow predicting model. And the results of cross-validation test (q 2 =0.9455) and method validation also showed the model of this study exhibited optimum stability and better predictive power than semiempirical method.