Increasingly, the restoration of large carnivores is proposed as a means through which to restore community structure and ecosystem function via trophic cascades. After a decades-long absence, African wild dogs (Lycaon pictus) recolonized the Laikipia Plateau in central Kenya, which we hypothesized would trigger a trophic cascade via suppression of their primary prey (dik-dik, Madoqua guentheri) and the subsequent relaxation of browsing pressure on trees. We tested the trophic-cascade hypothesis using (1) a 14-year time series of wild dog abundance; (2) surveys of dik-dik population densities conducted before and after wild dog recovery; and (3) two separate, replicated, herbivore-exclusion experiments initiated before and after wild dog recovery. The dik-dik population declined by 33% following wild dog recovery, which is best explained by wild dog predation. Dik-dik browsing suppressed tree abundance, but the strength of suppression did not differ between before and after wild dog recovery. Despite strong, top-down limitation between adjacent trophic levels (carnivore-herbivore and herbivore-plant), a trophic cascade did not occur, possibly because of a time lag in indirect effects, variation in rainfall, and foraging by herbivores other than dik-dik. Our ability to reject the trophic-cascade hypothesis required two important approaches: (1) temporally replicated herbivore exclusions, separately established before and after wild dog recovery; and (2) evaluating multiple drivers of variation in the abundance of dik-dik and trees. While the restoration of large carnivores is often a conservation priority, our results suggest that indirect effects are mediated by ecological context, and that trophic cascades are not a foregone conclusion of such recoveries.