Fire is considered a critical management tool in fire prone landscapes. Often studies and policies relating to fire focus on why and how the fire regime should be managed, often neglecting to subsequently evaluate management's ability to achieve these objectives over long temporal and large spatial scales. This study explores to what extent the long-term spatio-temporal fire patterns recorded in the Kruger National Park, South Africa has been influenced by management policies and to what extent it was dictated by underlying variability in the abiotic template. This was done using a spatially explicit fire-scar database from 1941 to 2006 across the 2 million ha Park. Fire extent (ha burnt per annum) 1) is correlated with rainfall cycles 2) exhibits no long-term trend and 3) is largely non-responsive to prevailing fire management policies. Rainfall, geology and distance from the closest perennial river and the interactions between these variables influence large-scale fire pattern heterogeneity: areas with higher rainfall, on basaltic substrates and far from rivers are more fire prone and have less heterogeneous fire regimes than areas with lower rainfall, on granitic substrates and closer to rivers. This study is the first to illustrate that under a range of rainfall and geological conditions, perennial rivers influence long-term, landscape-scale fire patterns well beyond the riparian zone (typically up to 15 km from the river). It was concluded that despite fire management policies which historically aimed for largely homogeneous fire return regimes, spatially and temporally heterogeneous patterns have emerged. This is primarily because of differences in rainfall, geology and distance from perennial rivers. We postulate that large-scale spatio-temporal fire pattern heterogeneity is implicit to heterogeneous savannas, even under largely homogenizing fire policies. Management should be informed by these patterns, embracing the natural heterogeneity-producing template. We therefore suggest that management actions will be better directed when operating at appropriate scales, nested within the broader implicit landscape patterns, and when focusing on fire regime parameters over which they have more influence (e.g. fire season).