1. Nutrient availability in terrestrial ecosystems has been found to vary along regional climatic and soil gradients and drive variation in plant community composition and vegetation structure. However, more local biotic feedbacks also affect nutrient availability, but their importance in determining vegetation structure relative to regional drivers is yet unclear. 2. Mesic African savannas form a transition zone between the dry grasslands with relatively low nitrogen availability (indicated by low plant N:P ratios) and the wet woodlands of the continent characterized by relatively low phosphorous availability (high plant N:P ratios). They host strong feedback mechanisms of both vegetation and consumers, where large grazers can create short grazing lawns that alternate over short distances with tall fire-dominated bunch grasslands, and patches dominated by woody species with abundant macrodetritivores. Here, we test if such local biotic interactions can overrule regional, landscape-level drivers of plant nutrient availability. 3. In a South African savanna, we find that plant N:P ratios, N and P resorption efficiencies and proficiencies, were indeed much stronger affected by small-scale biotic-created heterogeneity than by a regional rainfall gradient (530-830 mm yr À1 ). Furthermore, we observe that differences in N:P ratios between vegetation structural types are not caused by simple differences in inherent plant traits. 4. This suggests that availability of N and P to plants is strongly contingent on the local biotic interplay between different vegetation structural types and within-system feedbacks linked to each vegetation type. The regional rainfall gradient did affect nutrient availability, but mainly through its effect on the relative distribution of the different vegetation structural types thus setting the range of possible outcomes of biotic interactions. 5. Synthesis. Regional and global studies that model savanna carbon and nutrient cycling only on the basis of regional gradients in soil and climatic conditions may insufficiently capture the dominant ecosystem processes involved.