We implement mercury (Hg) biogeochemistry in the offline global 3-D ocean tracer model (OFFTRAC) to investigate the natural Hg cycle, prior to any anthropogenic input. The simulation includes three Hg tracers: dissolved elemental (Hg0aq), dissolved divalent (HgIIaq), and particle-bound mercury (HgPaq). Our Hg parameterization takes into account redox chemistry in ocean waters, air-sea exchange of Hg0, scavenging of HgIIaq onto sinking particles, and resupply of HgIIaq at depth by remineralization of sinking particles. Atmospheric boundary conditions are provided by a global simulation of the natural atmospheric Hg cycle in the GEOS-Chem model. In the surface ocean, the OFFTRAC model predicts global mean concentrations of 0.16 pM for total Hg, partitioned as 80% HgIIaq, 14% Hg0aq, and 6% HgPaq. Total Hg concentrations increase to 0.38 pM in the thermocline/intermediate waters (between the mixed layer and 1000 m depth) and 0.82 pM in deep waters (below 1000 m), reflecting removal of Hg from the surface to the subsurface ocean by particle sinking followed by remineralization at depth. Our model predicts that Hg concentrations in the deep North Pacific Ocean (>2000 m) are a factor of 2–3 higher than in the deep North Atlantic Ocean. This is the result of cumulative input of Hg from particle remineralization as deep waters transit from the North Atlantic to the North Pacific on their ~2000 year journey. The model is able to reproduce the relatively uniform concentrations of total Hg observed in the old deep waters of the North Pacific Ocean (observations: 1.2 ± 0.4 pM; model: 1.1 ± 0.04 pM) and Southern Ocean (observations: 1.1 ± 0.2 pM; model: 0.8 ± 0.02 pM). However, the modeled concentrations are factors of 5–6 too low compared to observed concentrations in the surface ocean and in the young water masses of the deep North Atlantic Ocean. This large underestimate for these regions implies a factor of 5–6 anthropogenic enhancement in Hg concentrations.