Coral and rocky reef fish populations are widely used as model systems for the experimental exploration of density-dependent vital rates, but patterns of density-dependent mortality in these systems are not yet fully understood. In particular, the paradigm for strong, directly density-dependent (DDD) postsettlement mortality stands in contrast to recent evidence for inversely density-dependent (IDD) mortality. We review the processes responsible for DDD and IDD per capita mortality in reef fishes, noting that the pattern observed depends on predator and prey behavior, the spatial configuration of the reef habitat, and the spatial and temporal scales of observation. Specifically, predators tend to produce DDD prey mortality at their characteristic spatial scale of foraging, but prey mortality is IDD at smaller spatial scales due to attack-abatement effects (e.g., risk dilution). As a result, DDD mortality may be more common than IDD mortality on patch reefs, which tend to constrain predator foraging to the same scale as prey aggregation, eliminating attack-abatement effects. Additionally, adjacent groups of prey on continuous reefs may share a subset of refuges, increasing per capita refuge availability and relaxing DDD mortality relative to prey on patch reefs, where the patch edge could prevent such refuge sharing. These hypotheses lead to a synthetic framework to predict expected mortality patterns for a variety of scenarios. For nonsocial, nonaggregating species and species that aggregate in order to take advantage of spatially clumped refuges, IDD mortality is possible but likely superseded by DDD refuge competition, especially on patch reefs. By contrast, for species that aggregate socially, mortality should be IDD at the scale of individual aggregations but DDD at larger scales. The results of nearly all prior reef fish studies fit within this framework, although additional work is needed to test many of the predicted outcomes. This synthesis reconciles some apparent contradictions in the recent reef fish literature and suggests the importance of accounting for the scale-sensitive details of predator and prey behavior in any study system.