Question: How do trees die in high-mortality and low-mortality Amazonian forest regions? Why do trees die in different ways? Location: Humid, lowland forests in Amazonian Peru and Venezuela. Methods: Patterns of multiple treefall and mode of death (standing, broken or uprooted) were recorded for trees ≥ 10 cm in diameter in permanent plots. Logistic regression was used to relate mode of death to tree diameter, relative growth rate and wood density. Results: Frequency of multiple death events was higher in high-mortality northwestern (NW) than in low-mortality northeastern (NE) Amazonia, but these events were small, averaging two trees killed per multiple death event. Breakage was the dominant known mode of death (51 ± 8%) in the NW, with half of fatal breakages caused by other treefalls or breakages. Small and slow-growing trees were more prone to breaking than uprooting. In NE Amazonia, the dominant known mode of death was standing (48 ± 10%); these trees tended to be relatively large and slow growing. Broken trees in NE forests have a lower wood density than uprooted trees. Conclusions: The major mortality mechanisms differ in the two regions. In the NW it involves an interaction between physiological failure and mechanical failure (small size, slow growth and broken mode). In the NE it is mainly driven by physiological failure (large size, slow growth and standing mode). We propose that by creating different-sized gaps the different dominant modes of death would favour species from different functional groups and so help to maintain the contrasting functional composition and mortality rates of the two regions. © 2009 International Association for Vegetation Science.