A number of large-scale sequestration strategies have been considered to help mitigate rising levels of atmospheric carbon dioxide (CO2). Here, we use an ocean general circulation model (OGCM) to evaluate the efficiency of one such strategy currently receiving much attention, the direct injection of liquid CO2 into selected regions of the abyssal ocean. We find that currents typically transport the injected plumes quite far before they are able to return to the surface and release CO2 through air sea gas exchange. When injected at sufficient depth (well within or below the main thermocline), most of the injected CO2 outgasses in high latitudes (mainly in the Southern Ocean) where vertical exchange is most favored. Virtually all OGCMs that have performed similar simulations confirm these global patterns, but regional differences are significant, leading efficiency estimates to vary widely among models even when identical protocols are followed. In this paper, we make a first attempt at reconciling some of these differences by performing a sensitivity analysis in one OGCM, the Princeton Modular Ocean Model. Using techniques we have developed to maintain both the modeled density structure and the absolute magnitude of the overturning circulation while varying important mixing parameters, we estimate the sensitivity of sequestration efficiency to the magnitude of vertical exchange within the low-latitude pycnocline. Combining these model results with available tracer data permits us to narrow the range of model behavior, which in turn places important constraints on sequestration efficiency. (C) 2004 Elsevier Ltd. All rights reserved.