Abstract Every year large proportions of northern Australia's tropical savanna landscapes are burnt, resulting in high fire frequencies and short intervals between fires. The dominant fire management paradigm in these regions is the use of low-intensity prescribed fire early in the dry season, to reduce the incidence of higher-intensity, more extensive wildfire later in the year. This use of frequent prescribed fire to mitigate against high-intensity wildfire has parallels with fire management in temperate forests of southern Australia. However, unlike in southern Australia, the ecological implications of high fire frequency have received little attention in the north. CSIRO and collaborators recently completed a landscape-scale fire experiment at Kapalga in Kakadu National Park, Northern Territory, Australia, and here we provide a synthesis of the effects of experimental fire regimes on biodiversity, with particular consideration of fire frequency and, more specifically, time-since-fire. Two recurring themes emerged from Kapalga. First, much of the savanna biota is remarkably resilient to fire, even of high intensity. Over the 5-year experimental period, the abundance of most invertebrate groups remained unaffected by fire treatment, as did the abundance of most vertebrate species, and we were unable to detect any effect of fire on floristic composition of the grass-layer. Riparian vegetation and associated stream biota, as well as small mammals, were notable exceptions to this general resilience. Second, the occurrence of fire, independent of its intensity, was often the major factor influencing fire-sensitive species. This was especially the case for extinction-prone small mammals, which have suffered serious population declines across northern Australia in recent decades. Results from Kapalga indicate that key components of the savanna biota of northern Australia favour habitat that has remained unburnt for at least several years. This raises a serious conservation concern, given that very little relatively long unburnt habitat currently occurs in conservation reserves, with most sites being burnt at least once every 2 years. We propose a conservation objective of increasing the area that remains relatively long unburnt. This could be achieved either by reducing the proportion of the landscape burnt each year, or by setting prescribed fires more strategically. The provision of appropriately long unburnt habitat is a conservation challenge for Australia's tropical savanna landscapes, just as it is for its temperate forests.