In this paper, we investigate which factors determine tree clustering in Southern African savannas. This was tested by measuring clustering of trees using the T-squared sampling method in plots of the Kruger National Park experimental burning programme in South Africa. Fire return interval is the main treatment in these plots, but also several auxiliary determining parameters like clay content in the soil, diameter of tree canopies, understorey composition, tree species diversity and average annual rainfall were measured while sampling. In the Kruger National Park 48 plots distributed over four different landscape types and with three different burning treatments (never, once every 3 y and annually) were sampled. First, we related the clustering of trees to these environmental variables. When looking at the most abundant species in each plot, the analysis revealed that clustering is mainly correlated with clay content in the soil. This analysis also showed that fire frequency had a positive effect on the clustering of tree species that are not very abundant. We suggest that less abundant species might be less resistant to fire and therefore adopt a mechanism of clustering to exclude grass fires under their canopy. Finally, we tested the effect of clustering on the impact of fire on trees by analysing the relationship between the distance of a tree to its nearest neighbour and its canopy diameter. We found that clustering reduces the damaging effect of fire on trees. Our study contributes to understanding of savanna functioning by showing which processes are relevant in the distribution of savanna trees.