Motivated by recent advances in mapping mesoscale eddy tracer mixing in the ocean we evaluate the sensitivity of a coarse-resolution global ocean model to a spatially variable neutral diffusion coefficient $\kappa_n(x,y,z)$. We gradually introduce physically-motivated models for the horizontal (mixing length theory) and vertical (surface mode theory) structure of $\kappa_n$ along with suppression of mixing by mean flows. Each structural feature influences the ocean’s hydrography and circulation to varying extents, with the suppression of mixing by mean flows being the most important factor and the vertical structure being relatively unimportant. When utilizing the full theory (experiment “FULL’) the interhemispheric overturning cell is strengthened by $2$ Sv at $26^\circ$N (a $\sim20\%$ increase), bringing it into better agreement with observations. Zonal mean tracer biases are also reduced in FULL. Neutral diffusion impacts circulation through surface temperature-induced changes in surface buoyancy fluxes and non-linear equation of state effects. Surface buoyancy forcing anomalies are largest in the Southern Ocean where decreased neutral diffusion in FULL leads to surface cooling and enhanced dense-to-light surface watermass transformation, reinforced by reductions in cabbeling and thermobaricity. The increased watermass transformation leads to enhanced mid-latitude stratification and interhemispheric overturning. The spatial structure for $\kappa_n$ in FULL is important as it enhances the interhemispheric cell without degrading the Antarctic bottom water cell, unlike a spatially-uniform reduction in $\kappa_n$. These results highlight the sensitivity of modeled circulation to $\kappa_n$ and motivate the use of physics-based models for its structure.