The large-scale mean circulation of the Atlantic Ocean is examined using a general circulation model (GCM) and its approximate adjoint. A cost function is specified that requires the model inputs to be consistent with hydrographic observations and observed air-sea fluxes of heat and freshwater, whereas the velocity field has to adjust to the modified thermohaline initial and boundary conditions. The optimized, quasi-steady model state is closer to the observed circulation state than previous prognostic steady-state models of comparable resolution. However, it is only partially consistent with the error estimates derived from the observations. In the western boundary region large deviations of the optimized surface fluxes from the observations occur. Additionally, the heat release of the ocean shows unrealistically high values at around 60N. At quasi-equilibrium, in large parts of the thermocline values for temperature and salinity along the Gulf Stream and the North Atlantic current are significantly lower than those in the hydrographic data, thus tending toward winter-time conditions. The model produces a meridional overturning cell with maximum values of 16 to 23 Sv for different experiments. The corresponding heat transports reach maximum values between 0.83 and 1.07 × 105 W. Model deficiencies like the inappropriate spatial and temporal resolution obviously prevent realistic estimates of water mass distributions and surface fluxes not only in the area of the western boundary current. Another shortcoming of the presented results is the parameterization of eddies and subgrid-scale processes by poorly known diffusion coefficients. To overcome these problems at least partly, future models based on the adjoint method should have a seasonal and increased spatial resolution.