Despite the large spatial extent and the obvious importance of the savanna biome, not to mention several decades of savanna research, the origin, age, nature, and dynamics of savannas are not well understood. Basically, the question surrounding the presence or existence of savannas focuses on the long-term coexistence of the dominating life forms – trees and grasses. How do these two very different components coexist, without one of them dominating the other, and what mechanisms determine the proportion of each? Earlier equilibrium concepts have recently been replaced by non-equilibrium concepts, and the current view is that tree-grass interactions in savannas cannot be predicted by a simple model. Instead, many interacting factors operating at various spatial and temporal scales contribute to creating and maintaining savanna physiognomy. In this paper we analyse a number of studies from savannas in different parts of the world and discuss whether a general pattern can be perceived behind the numerous factors influencing the presence of savannas systems. On the basis of this analysis we propose a new unifying concept of savanna existence, i.e., the concept of ecological buffering mechanisms. In contrast to previous approaches to explain tree-grass coexistence in savannas, the concept of buffering mechanisms does not focus on equilibria or non-equilibria, steady states of the system or domains of attraction. Instead, in the concept of ecological buffering mechanisms we suggest that it is much more useful to focus on the boundaries of savanna existence itself and to investigate the mechanisms that allow a savanna to persist in critical situations where this system is driven to its boundaries, e.g., pure grasslands or tropical forests. The concept of ecological buffering mechanisms integrates both earlier concepts of ecological theory and general ideas on savanna dynamics as well as specific studies of savannas in different parts of the world.