Abstract Fire is a dominant feature of tropical savannas throughout the world, and provides a unique opportunity for habitat management at the landscape scale. We provide the background and methodology for a landscape-scale savanna fire experiment at Kapalga, located in Kakadu National Park in the seasonal tropics of northern Australia. The experiment addresses the limitations of previous savanna fire experiments, including inappropriately small sizes of experimental units, lack of replication, consideration of a narrow range of ecological responses and an absence of detailed measurement of fire behaviour. In contrast to those elsewhere in the world, Australia's savannas are sparsely populated and largely uncleared, with fires lit primarily in a conservation, rather than pastoral, context. Fire management has played an integral role in the traditional lifestyles of Aboriginal people, who have occupied the land for perhaps 50 000 years or more. Currently the dominant fire management paradigm is one of extensive prescribed burning early in the dry season (May-June), in order to limit the extent and severity of fires occurring later in the year. The ecological effects of different fire regimes are hotly debated, but we identify geo-chemical cycling, tree demography, faunal diversity and composition, phenology, and the relative importance of fire intensity, timing and frequency, as critical issues. Experimental units (‘compartments’) at Kapalga are 15–20km2 catchments, centred on seasonal creeks that drain into major rivers. Each compartment has been burnt according to one of four treatments, each replicated at least three times: ‘Early’- fires lit early in the dry season, which is the predominant management regime in the region; ‘Late’- fires lit late in the dry season, as occurs extensively in the region as unmanaged ‘wildfires’; ‘Progressive’- fires lit progressively throughout the dry season, such that different parts of the landscape are burnt as they progressively dry out (believed to approximate traditional Aboriginal burning practices); and ‘Unburnt’- no fires lit, and wildfires excluded. All burning treatments have been applied annually for 5 years, from 1990 to 1994. Six core projects have been conducted within the experimental framework, focusing on nutrients and atmospheric chemistry, temporary streams, vegetation, insects, small mammals, and vertebrate predators. Detailed measurements of fire intensity have been taken to help interpret ecological responses. The Kapalga fire experiment is multidisciplinary, treatments have been applied at a landscape scale with replication, and ecological responses can be related directly to measurements of fire intensity. We are confident that this experiment will yield important insights into the fire ecology of tropical savannas, and will make a valuable contribution to their conservation management.