A critically evaluated set of 94 soil water partition coefficients normalized to soil organic carbon content (K_oc) is presented for 11 classes of organic chemicals. This data set is used to develop and evaluate K_oc estimation methods using three different descriptors. The three types of descriptors used in predicting K_oc were octanol/water partition coefficient (K_ow), molecular connectivity (^mX_t) and linear solvation energy relationships (LSERs). The best results were obtained estimating K_oc from K_ow, though a slight improvement in the correlation coefficient was obtained by using a two-parameter regression with K_ow and the third order difference term from ^mX_t. Molecular connectivity correlations seemed to be best suited for use with specific chemical classes. The LSER provided a better fit than ^mX_t but not as good as the correlation with K_oc. The correlation to predict K_oc from K_ow was developed for 72 chemicals; log K_oc = 0.903* log K_ow + 0.094. This correlation accounts for 91% of the variability in the data for chemicals with log K_ow ranging from 1.7 to 7.0. The expression to determine the 95% confidence interval on the estimated K_oc is provided along with an example for two chemicals of different hydrophobicity showing the confidence interval of the retardation factor determined from the estimated K_oc. The data showed that K_oc is not likely to be applicable for chemicals with log K_ow < 1.7. Finally, the K_oc correlation developed using K_ow as a descriptor was compared with three nonclass-specific correlations and two 'commonly used' class-specific correlations to determine which method(s) are most suitable.