A three-dimensional ocean model with an idealized geometry and coarse resolution coupled to a two-dimensional (zonally-averaged) statistical-dynamical atmospheric model is used to simulate the response of the thermohaline circulation to increasing CO2 concentration in the atmosphere. The relative role of different factors in slowing down the thermohaline circulation was studied by performing simulations with ocean only and partially coupled models. The computational efficiency of the model allows an extensive and thorough study of the causes of changes in the strength of the thermohaline circulation, through a large number of extended runs. The increase in the atmosphere-to-ocean surface heat fluxes is shown to be the dominant factor in both causing the weakening of the circulation in response to an increasing external forcing as well as in controlling the subsequent recovery. Changes in the zonal distribution of heat fluxes serve as a positive feedback for both decrease and recovery of the meridional overturning, and turn out to be as important as changes in the zonal-mean values of heat fluxes. We also demonstrate that the recovery of the circulation in the ocean model cannot be sustained without feedbacks from the atmosphere. The dependency of global and regional responses on parameterization of eddy mixing, namely the Gent-McWilliams parameterization scheme versus horizontal diffusion, is also discussed.