A stochastic biogeochemical model has been developed to simulate the transformation and degradation of natural organic matter (NOM) using an agent-based algorithm which treats each molecule as a separate and potentially unique entity. Molecules react when a pseudo-random number is lower than the calculated reaction probability in a given time step; repeated time steps simulate the transformation of precursor molecules into a complex NOM assemblage. The data for each molecule-elemental and functional group composition-can be used to calculate many properties directly and exactly for each molecule in the assemblage, e.g., molecular weight (MW), fraction of aromatic C (Ar), and charge at pH 7 (Z). Empirical quantitative structure activity relationships (QSARs) are developed which permit the estimation of thermodynamic quantities K ow (the octanol-water partition coefficient) and pK a (acidity) for each molecule. Root mean square errors for these QSARs are 0.39 log units for log K ow and 0.45 log units for pK a . Distributions of both exactly calculated (MW, Ar, Z) and estimated thermodynamic (K ow , pK a ) properties are examined and compared with published experimental data. Molecular weight distributions from size exclusion HPLC experiments on aquatic NOM are quantitatively similar to simulation results. pH titrations and polarity distributions from reversed-phase HPLC are qualitatively similar to simulation results. This agreement suggests that the agent-based model can be used to explore hypotheses regarding both compositional and thermodynamic properties of NOM.Keywords Agent-based model Á Humic Á Natural organic matter Á Molecular weight Á Aromaticity Á Acidity Á Polarity Á QSAR Abbreviations a i constant value in a linear equation amu atomic mass units Ar aromaticity, the fraction of non-carbonyl sp 2 C in a molecule or group of molecules Robert Wetzel-deceased.