Abstract The impact of feral Asian water buffalo (Bubalus bubalis) and season of fire on growth and survival of mature trees was monitored over 8 years in the eucalypt savannas of Kakadu National Park. Permanently marked plots were paired on either side of a 25-km-long buffalo-proof fence at three locations on an elevational gradient, from ridge-top to the edge of a floodplain; buffalo were removed from one side of the fence. All 750 trees ≥ 1.4 m height were permanently marked; survival and diameter of each tree was measured annually; 26 species were grouped into four eco-taxonomic groups. The buffalo experiment was maintained for 7 years; trees were monitored an additional year. Fires were excluded from all sites the first 3 years, allowed to occur opportunistically for 4 years and excluded for the final year. Fires were of two main types: low-intensity early dry season and high-intensity late dry season. Growth rates of trees were size-specific and positively related to diameters as exponential functions; trees grew slowest on the two ends of the gradient. Eucalypt mortality rates were 1.5 and 3 times lower than those of pantropics and of arborescent monocots, respectively, but the relative advantage was lost with fires or buffalo grazing. Without buffalo grazing, ground level biomass was 5–8 t ha−1 compared with 2–3 t ha−1, within 3 years. In buffalo-absent plots, trees grew significantly slower on the dry ridge and slope, and had higher mortality across the entire gradient, compared with trees in buffalo-present plots. At the floodplain margin, mortality of small palms was higher in buffalo-present sites, most likely due to associated heavy infestations of weeds. Low-intensity fires produced tree growth and mortality values similar to no-fire, in general, but, like buffalo, provided a ‘fertilization’ effect for Eucalyptus miniata and Eucalyptus tetrodonta, increasing growth in all size classes. High-intensity fires reduced growth and increased mortality of all functional groups, especially the smallest and largest (>35 cm d.b.h.) trees. When buffalo and fires were excluded in the final year, there were no differences in growth or mortality between paired sites across the environmental gradient. After 8 years, the total numbers of trees in buffalo-absent plots were only 80% of the number in buffalo-present plots, due to relatively greater recruitment of new trees in buffalo-present plots; fire-sensitive pantropics were particularly disadvantaged. Since the removal of buffalo is disadvantageous, at least over the first years, to savanna tree growth and survival due to a rebound effect of the ground-level vegetation and subsequent changes in fire-vegetation interactions, process-orientated management aimed at reducing fuel loads and competitive pressure may be required in order to return the system to a previous state. The ‘footprint’ of 30 years of heavy grazing by buffalo has implications for the interpretation of previous studies on fire-vegetation dynamics and for current research on vegetation change in these savannas.