The divergence of the eddy mass flux in the surface layer of the Southern Ocean makes an important contribution to subduction of fluid through the base of the mixed layer. Therefore, accurate parameterization of this process is needed to correctly represent the Southern Ocean ventilation in coarse-resolution models. We test a common approach to the parameterization of eddy fluxes (Gent and McWilliams, 1990) using output from the 1/6° eddy-permitting Southern Ocean State Estimate, which assimilates a variety of ocean observations using an adjoint method. When a constant diffusion coefficient of conventional magnitude O(1000 m2 s−1) is used, the parameterized fluxes fail to reproduce the regional pattern and magnitude of eddy-driven subduction diagnosed from the model. However, when an appropriate choice is made for the diffusion coefficient, the parameterization does a good job of reproducing the distribution and strength of the eddy contribution to subduction. Using a spatially-varying coefficient is key to reproduce the regional pattern of the eddy-induced subduction. In addition, the magnitude of the subduction is correctly represented only with a diffusion coefficient that peaks at 104 m2 s−1 in the most energetic areas of the Southern Ocean, a factor of ten larger than commonly used in coarse-resolution climate models. Using a diffusion coefficient that is too small will underestimate the contribution of eddies to the ocean sequestration of heat, salt and carbon.