AbstractSavannas are spatially diverse, variable and are susceptible to high rates of disturbance from fire and herbivory. There is significant interest in woody cover dynamics in relation to disturbance regimes. Less effort has been devoted to understand processes that drive tree community composition. In this study, tree species composition data collected at the landscape scale in the Serengeti were used to identify key environmental factors driving variation in species composition. A system of 38 plots clustered within 10 sites spanning the mean annual precipitation (MAP) gradient was used to assess the relative role of bottom‐up (precipitation, soil nutrients and soil texture) vs. top‐down factors (fire and elephant herbivory) on tree community composition. We developed candidate models relating tree species composition (based on multivariate community analysis) to different combinations of plot‐level environmental covariates. Results suggest that tree community composition is largely driven by MAP and is associated with elephant population density. Strikingly, we found no evidence that fire influences species compositional turnover. In a second analysis, we used structural equation model (SEM) to explore the possible direction of association between elephant density and tree species composition. We compared a model that included elephant effects on composition to one that included community composition effects on elephant density. Results suggest that variation in elephant population density across space is more likely to drive tree community composition and not vice versa. We propose that precipitation and herbivory, rather than fire, determine tree species composition in Serengeti Acacia tree community.