Large-mammal populations are ecological linchpins, and their worldwide decline and extinction disrupts many ecosystem functions and services. Reversal of this trend will require an understanding of the determinants of population decline, to enable more accurate predictions of when and where collapses will occur and to guide the development of effective conservation and restoration policies. Many correlates of large-mammal declines are known, including low reproductive rates, overhunting, and habitat destruction. However, persistent uncertainty about the effects of one widespread factor-armed conflict-complicates conservation-planning and priority-setting efforts. Case studies have revealed that conflict can have either positive or negative local impacts on wildlife, but the direction and magnitude of its net effect over large spatiotemporal scales have not previously been quantified. Here we show that conflict frequency predicts the occurrence and severity of population declines among wild large herbivores in African protected areas from 1946 to 2010. Conflict was extensive during this period, occurring in 71% of protected areas, and conflict frequency was the single most important predictor of wildlife population trends among the variables that we analysed. Population trajectories were stable in peacetime, fell significantly below replacement with only slight increases in conflict frequency (one conflict-year per two-to-five decades), and were almost invariably negative in high-conflict sites, both in the full 65-year dataset and in an analysis restricted to recent decades (1989-2010). Yet total population collapse was infrequent, indicating that war-torn faunas can often recover. Human population density was also correlated (positively) with wildlife population trajectories in recent years; however, we found no significant effect, in either timespan, of species body mass, protected-area size, conflict intensity (human fatalities), drought frequency, presence of extractable mineral resources, or various metrics of development and governance. Our results suggest that sustained conservation activity in conflict zones-and rapid interventions following ceasefires-may help to save many at-risk populations and species.