We present two laboratory experiments designed to show how cannibalism and its effects on population numbers and size structure vary in response to density, size, and food manipulations in larvae of the dragonfly Epitheca cynosura. Because young larvae of this species interact at very high densities after hatching asynchronously from clumps of egg masses, our efforts focused on these early stages. In one experiment, we varied the combination of sizes (instars) and the presence of food for 233 pairs of early-instar larvae. Cannibalism was uncommon if larvae were of the same instar (only 2.4% of such pairs exhibited cannibalism), frequent (53%) if larvae differed by one instar, and certain (100%) if a two-instar size difference was present. The rate at which cannibalism increased with size differences between larvae was greater when food was absent than it was when food was provided. In a second experiment, we subjected replicate cohorts of newly hatched dragonflies to manipulations of hatching synchrony, initial density, and food availability. Asynchronous hatching over 25 d produced a broad size distribution apparently conducive to cannibalism, whereas synchronous hatching over 3 d initially precluded cannibalism. As a result, strong density-dependent mortality only appeared following the asynchronous hatch. For high-density, asynchronous treatments, cannibalism significantly reduced size variation and tightened size distributions over the course of development. Survivors from high-density treatments were significantly larger than were low-density survivors. More abundant food allowed larvae to reach larger sizes, but did not improve survival. Survival was affected mainly by the density and size distribution of larvae. These results suggest that cannibalism was more important than exploitative competition for food in determining the size and survival of dragonflies in the laboratory. We conclude that when juveniles hatch asynchronously in close proximity, cannibalism can: (1) contribute to population regulation by imposing greater per capita mortality at high densities, and (2) increase population synchrony by exerting size-specific mortality on smaller individuals throughout development.