Numerical experiments with a medium-resolution primitive equation model of the South Atlantic mean circulation are described. The results from the standard model realization indicate that the model succeeds in representing the large-scale transport and circulation features. However, a comparison with a velocity field derived from surface drifter data reveals discrepancies of the modeled velocities from the observations in magnitude as well as direction of the flow field. In order to diminish the model deviations from the data, an attempt is made to couple the model to the observations through a simple data assimilation technique. The assimilated model succeeds in improving the subtropical gyre circulation. Only a minor effect on the basin-scale integrated quantities is observed. However, the density field may be deformed as a response to the assimilation of velocity data without simultaneously adapting a corresponding density structure. The influence of the disturbance of the density structure is most prominent at the edges of the observed data set, which does not cover the entire model domain, and is confined to the upper ocean and balanced above the thermocline. We calculated a meridional heat transport that is generally in accordance with estimates from other sources. The analysis of heat and salt fluxes suggests that the model features both the so-called warm water path and cold water path in closing the global thermohaline circulation. While heat is mainly imported in surface and thermocline waters with the Agulhas Current around South Africa, it is the Antarctic Intermediate Water that compensates for more than 50% of the salt loss by the outflowing North Atlantic Deep Water. ¿ 1998 American Geophysical Union