Using a general theory for partition coefficients based on a quantum chemically derived conductor-like screening model for real solvents sigma-moment descriptors, the logarithmic soil sorption coefficients log K(oc) of a database of 440 compounds has been successfully correlated, achieving a standard deviation (root-means-squared [RMS]) of 0.62 log-units on the training set and a predictive RMS of 0.72 log-units on a more demanding test set. The quality of this generally applicable predictive approach is almost the same as that of a regression of log K(oc) with experimental log K(ow) values, which are the best correlations currently available. The error of this new predictive method is only approximately 43% of the error of a recently published model using a different quantum chemically based approach.