The relative importance of intraspecific, interspecific, and seasonal causes of larval mortality were investigated for aquatic larvae of the giant damselfly Megaloprepus coerulatus in Panama. These larvae live in water-filled holes in fallen and living trees, where they and three other common odonate species are the top predators. By mid wet season, M. coerulatus larvae were found in nearly half of all tree holes that harbored odonates. Although M. coerulatus were typically, but not always, eliminated from holes inhabited by larger hetero-specifics, M. coerulatus were more likely to encounter conspecifics than other odonate species. Hole with less than 11 of water rarely contained more than a single larva. In large holes where M. coerulatus was the only odonate species present, multiple larvae coexisted at a density of one larva per 1–21 of water. There the absence of 2–4 of the 5 larval size classes, despite a continuous input of eggs, suggested that cannibalism was a common cause of mortality. The size of the final instar, which determined adult body size, was correlated positively with tree hole volume for male, but not female, larvae. Experiments showed that when two larvae were placed together in 0.4–1 holes with abundant tadpole prey, the larger larva killed the smaller one. Often the larva that was killed was not eaten. Small larvae were more tolerant of each other than were pairs of medium or large larvae. Before killing occurred, the presence of larger larvae reduced the growth of smaller individuals, relative to controls. ‘Obligate’ killing was density-dependent. In 3.0–1 holes with ad libitum prey, conspecific killing occurred until the larval density stabilized at one larva per 1–1.5 I, similar to the density found in large holes under field conditions, For M. coerulatus, cannibalism functions to reduce the number of potential competitors for food in addition to providing nutrition. When interactions between paired larvae in small holes were experimentally prevented, competition for food reduced the growth of one or both larvae relative to controls. Holes that were watered during the dry season supported larval densities similar to those in the wet season. Thus, dry season mortality could not be attributed to a decrease in available prey. Rather, M. coerulatus larvae could not survive more than 1 month of complete drying. Because the dry season typically lasts more than 6 weeks, habitat drying is a secondary source of mortality, affecting second- or third-generation larvae that fail to emerge before tree holes dry out completely.