The central Kalahari of Botswana straddles the major ecotone between the broad-leafed and microphyllous (fine-leafed) savanna communities. Except in the immediate vicinity of the relatively small number of permanent settlements, this extensive tract of land is sparsely inhabited, and thus ideal for evaluating various determinants of savanna structure under conditions of minimal human disturbance. This study summarizes textural and mineralogical data for the central Kalahari cover sands, and links these to vegetation distribution patterns. The overall distribution of woody species within the central Kalahari is consistent with precipitation as the dominant determinant, but also shows a significant correlation with sand grain size and heavy-mineral content. However, sand grain size and heavy-mineral content per se do not play a major role in influencing species distribution. Both increase towards the margin of the Kalahari basin, reflecting shallower bedrock, and it is inferred that depth to bedrock exerts a secondary influence on species distribution near the extremes of their rainfall-defined distribution ranges. With shallower bedrock, deeper-rooting species, able to exploit sub-Kalahari aquifers, will displace those with well-developed lateral root systems, adapted to exploit near-surface water. The available evidence indicates that soil nutrient content does not have a major effect on vegetation communities in the sandveld environment of the central Kalahari. However, further work is required to substantiate this conclusion. There is also a significant correlation between savanna structure and grain size. Extremely fine-grained soils favour a shrub savanna, while tree savannas are associated with coarser-grained soils. This is consistent with deeper penetration of moisture in the latter, in line with classical models for savanna structure. However, an alternative explanation for this relationship is that variations in grain size mirror changes in nutrient content, which is the ultimate control on vegetation structure. Further work is also needed to test these contrasting models.