Variation in the vulnerability of herbivore prey to predation is linked to body size, yet whether this relationship is size‐nested or size‐partitioned remains debated. If size‐partitioned, predators would be focused on prey within their preferred prey size range. If size‐nested, smaller prey species should become increasingly more vulnerable because increasingly more predators are capable of catching them. Yet, whether either of these strategies manifests in top–down prey population limitation would depend both on the number of potential predator species as well as the total mortality imposed. Here we use a rare ecosystem scale ‘natural experiment’ comparing prey population dynamics between a period of intense predator persecution and hence low predator densities and a period of active predator protection and population recovery. We use three decades of data on herbivore abundance and distribution to test the role of predation as a mechanism of population limitation among prey species that vary widely in body size. Notably, we test this within one of the few remaining systems where a near‐full suite of megaherbivores occur in high density and are thus able to include a thirtyfold range in herbivore body size gradient. We test whether top–down limitation on prey species of particular body size leads to compositional shifts in the mammalian herbivore community. Our results support both size‐nested and size‐partitioning predation but suggest that the relative top–down limiting impact on prey populations may be more severe for intermediate sized species, despite having fewer predators than small species. In addition we show that the gradual recovery of predator populations shifted the herbivore community assemblage towards large‐bodied species and has led to a community that is strongly dominated by large herbivore biomass.