The soil sorption coefficient, K(OC), is an important property affecting the environmental fate of organic molecules. Difficulties associated with measuring K(OC) have led to many attempts to predict this property, but most rely on empirical descriptors for the soil phase determined from correlations with measured K(OC) data, and are thereby limited by the data quality and diversity. A new method is presented to predict K(OC) for nonionic organic compounds that requires only molecular structures. No calibration is performed. Using model humic acid (HA) and fulvic acid (FA) molecular structures from the literature, the soil organic matter is modeled as an organic solvent composed of HA or FA molecules. K(OC) is predicted as an organic solvent-water partition coefficient using the quantum mechanics-based model COSMO-SAC. The log K(OC) values for a set of 440 diverse, environmentally relevant chemicals are predicted with a root-mean-square error of 0.84-1.08, depending on which model HA or FA is used.