The savanna vegetation of southern tropical Africa is characterized by co-dominance of C4 grasslands and C3 woodlands. Long-term variations in the tropical savanna vegetation in arid and semi-arid climates are commonly considered to be primarily sensitive to precipitation and atmospheric CO2 concentrations. The sensitivity of tropical vegetation to temperature, however, is often considered as secondary or negligible, particularly in paleostudies due to difficulties of reconstructing terrestrial temperature in the tropics. In this study, we use the terrestrial vegetation model BIOME4, which was forced by climate simulations from the Kiel Climate Model (KCM) for the Holocene and by climate reconstructions for the most recent glacial period to understand reconstructed vegetation changes in southern tropical Africa of the past 37,000 yr. We focus on these two periods because vegetation reconstructions from a marine sediment core near the Zambezi River mouth cannot be explained by precipitation changes and changes of atmospheric CO2 alone. For the Holocene, we force BIOME4 simulations with reconstructed atmospheric CO2 concentrations, and spatial and seasonal climate patterns from the early- and mid-Holocene (9.5 and 6 ka BP) simulations with the KCM. For the glacial period, we analyze idealized experiments based upon reconstructed temperature, precipitation and CO2 at 31, 28 and 21 ka BP. Our study shows that both Holocene and glacial simulations of vegetation cover exhibit good agreement with reconstructed C4:C3C4:C3 ratios when temperature changes are taken into account. While both precipitation and temperature control the C4:C3C4:C3 ratio during the Holocene atmospheric CO2 and temperature variations are major factors controlling vegetation changes during the glacial period. In our simulations, variations in temperature along with precipitation and atmospheric CO2 reconcile the evolution of vegetation observed in the Zambezi catchment during the last 37,000 yr. In consequence, the effect of temperature variations on tropical savanna vegetation should be taken into account with respect to modeling past or future climates.