Anthropogenic understory fires affect large areas of tropical forest, particularly during severe droughts. Yet, the mechanisms that control tropical forests' susceptibility to fire remain ambiguous. We tested the widely accepted hypothesis that Amazon forest fires increase susceptibility to further burning by conducting a 150 ha fire experiment in a closed-canopy forest near the southeastern Amazon forest–savanna boundary. Forest flammability and its possible determinants were measured in adjacent 50 ha forest plots that were burned annually for 3 consecutive years (B3), once (B1), and not at all (B0). Contrary to expectation, an annual burning regime led to a decline in forest flammability during the third burn. Microclimate conditions were more favorable compared with the first burn (i.e. vapor pressure deficit increased and litter moisture decreased), yet flame heights declined and burned area halved. A slight decline in fine fuels after the second burn appears to have limited fire spread and intensity. Supporting this conclusion, fire spread rates doubled and burned area increased fivefold in B3 subplots that received fine fuel additions. Slow replacement of surface fine fuels in this forest may be explained by (i) low leaf litter production (4.3 Mg ha−1 yr−1), half that of other Amazon forests; and (ii) low fire-induced tree and liana mortality (5.5±0.5% yr−1, SE, in B3), the lowest measured in closed-canopy Amazonian forests. In this transitional forest, where severe seasonal drought removed moisture constraints on fire propagation, a lack of fine fuels inhibited the intensity and spread of recurrent fire in a negative feedback. This reduction in flammability, however, may be short-lived if delayed tree mortality or treefall increases surface fuels in future years. This study highlights that understanding fuel input rate and timing relative to fire frequency is fundamental to predicting transitional forest flammability – which has important implications for carbon emissions and potential replacement by scrub vegetation.