Daily averaged positions obtained from satellite-tracked drifting buoys launched during the First GARP Global Experiment (FGGE) and those launched as part of the ONR sponsored Southern Ocean Studies (SOS) are used to analyze eddy diffusivity due to mesoscale motions and its parameterization in some regions of the South Atlantic Ocean, significantly extending the energy range for which this type of estimations are available. Diffusion coefficients, Lagrangian integral time scales and velocity variances are calculated, and Taylor's (1921) Theorem which relates eddy diffusivity estimates with the dispersion of particles from a fixed origin is tested. In all of the cases analyzed here, the anisotropy of the statistics appears as an outstanding feature, with zonal values being larger than meridional ones. Although the conditions of stationarity and homogeneity of the statistics required by Taylor's Theorem are not fulfilled, a positive test for relation was observed for small lags. However, for later times, diffusion and dispersion curves have a tendency to diverge due to the influence of mean current shear. This interpretation is evident in the large differences between diffusion and dispersion in the Brazil/Malvinas extension where there is substantial current shear and a very close agreement between the curves in the gyre interior where the shear is small. The relationship between the diffusion coefficients and eddy velocity statistics for different regimes in the gyre are explored with the conclusion that it is impossible to decide between the hypothesis that the diffusion is proportional to the eddy velocity, i.e. mixing-length theory applies, and the alternate case of the diffusion scaling with the velocity variance or eddy energy. The implications this has to any global parameterization of eddy diffusion is discussed.