In addition to analytical speciation, reliable Hg species modeling is crucial for predicting the mobility and toxicity of Hg, but geochemical speciation codes have not yet been tested for their prediction accuracy. Our study compares analyses of Hg species in highly Hg-contaminated groundwater (Hgtot: 0.02-4 μmol·L(-1)) at three sites with predictions of Hg speciation obtained from three geochemical codes (WHAM, Visual MINTEQ, PHREEQC) with and without implementation of Hg complexation by dissolved organic matter (DOM). Samples were analyzed for chemical composition, elemental, inorganic, and DOM-bound Hg (Hg(0), Hginorg, HgDOM). Hg-DOM complexation was modeled using three approaches: binding to humic/fulvic acids, binding to thiol-groups, or a combination of both. Results of Hg(0) modeling were poor in all scenarios. Prediction accuracy for Hginorg and HgDOM strongly depended on the assumed DOM composition. Best results were achieved when weaker binding sites, simulated by WHAMs DOM submodel, were combined with strongly binding thiol groups. Indications were found that thiol-DOM ratios in groundwater are likely to be lower than in surface water, highlighting the need for analytical thiol quantification in groundwater DOM. This study shows that DOM quality is a crucial parameter for prediction of Hg speciation in groundwater by means of geochemical modeling.