Anthropogenic understory fires have affected large areas of tropical forest in recent decades particularly during severe droughts. Yet the mechanisms that control fire-induced mortality of tropical trees and lianas remain ambiguous due to the challenges associated with documenting mortality given variation in fire behavior and forest heterogeneity. In a seasonally dry Amazon forest we conducted a burn experiment to quantify how increasing understory fires alter patterns of stem mortality. From 2004 to 2007 tree and liana mortality was measured in adjacent 50-ha plots that were intact (B0 – control) burned once (B1) and burned annually for 3 years (B3). After 3 years cumulative tree and liana mortality (=1 cm dbh) in the B1 (5.8% yr-1) and B3 (7.0% yr-1) plots significantly exceeded mortality in the control (3.2% yr-1). However these fire-induced mortality rates are substantially lower than those reported from more humid Amazonian forests. Small stems were highly vulnerable to fire-induced death contrasting with drought-induced mortality (measured in other studies) that increases with tree size. For example one low-intensity burn killed >50% of stems <10 cm within a year. Independent of stem size species-specific mortality rates varied substantially from 0% to 17% yr-1 in the control 0% to 26% yr-1 in B1 and 1% to 23% yr-1 in B3 with several species displaying high variation in their vulnerability to fire-induced mortality. Protium guianense (Burseraceae) exhibited the highest fire-induced mortality rates in B1 and B3 which were 10- and 9-fold greater than the baseline rate. In contrast Aspidosperma excelsum (Apocynaceae) appeared relatively unaffected by fire (0.3% to 1.0% mortality yr-1 across plots) which may be explained by fenestration that protects the inner concave trunk portions from fire. For stems =10 cm both char height (approximating fire intensity) and number of successive burns were significant predictors of fire-induced mortality whereas only the number of consecutive annual burns was a strong predictor for stems <10 cm. Three years after the initial burn 62 ± 26 Mg ha-1 (s.e.) of live biomass predominantly stems <30 cm was transferred to the dead biomass pool compared with 8 ± 3 Mg ha-1 in the control. This biomass loss from fire represents ~30% of this forest\\\s aboveground live biomass (192 (±3) Mg ha-1; >1 cm DBH). Although forest transition to savanna has been predicted based on future climate scenarios our results indicate that wildfires from agricultural expansion pose a more immediate threat to the current carbon stocks in Amazonian forests.