A critical examination of data, assumptions and statistical approaches for previously published QSARs for the octanol-water and soil-water partition coefficients (K OW and K OC) of polychlorinated biphenyl (PCBP) congeners identified a number of error sources. A variety of new QSARs were developed based on weighted homogeneity of slopes models. The best type of models used the degree of chlorination in the orthoposition of the PCB molecule as predictor variable together with either the calculated molecular total surface area (TSA) or the gas chromatographic relative retention times (RRT). Prediction of logK OW and logK OC for all 209 PCB congeners were based on 53 and 48 experimental data points, respectively. The predicted values were compared to test data sets of 12 and 27 experimental data points, respectively. The QSAR models gave predictions in good agreement with experimental test data.