1 Savanna ecosystems are characterized by the codominance of two different life forms: grasses and trees. An operational understanding of how these two different life forms coexist is essential for understanding savanna function and for predicting its response to future environmental change.

2 The existing model, which proposes that grasses and trees coexist by a separation of rooting niches, is not supported by recent empirical investigations. Our aim was to define an alternative mechanism of grass–tree coexistence in savanna ecosystems. The model we have built concentrates on life history–disturbance interactions between grasses and trees.

3 The model demonstrates coexistence for a wide range of environmental conditions, and exhibits long periods of slow decline in adult tree numbers interspersed with relatively infrequent recruitment events. Recruitment is controlled by rainfall, which limits seedling establishment, and fire, which prevents recruitment into adult size classes. Decline in adult tree numbers is the result of continuing, but low levels, of adult mortality. Both aspects of the dynamics are consistent with an established non-equilibrium mechanism of coexistence (the storage effect).

4 A sensitivity analysis indicated that data on tree resprouting ability, stem growth rates and the relationship between seedling establishment and wet season drought are essential for predicting both the range of conditions for which coexistence is possible and the response of savanna ecosystems to environmental change.

5 Our analysis suggests that understanding grass–tree interactions in savanna requires consideration of the long-term effects of life history–disturbance interactions on demography, rather than the fine-scale effects of resource competition on physiological performance.