A first analysis of the stability of trophic structure following tropical forest fragmentation was performed in an experimentally fragmented tropical forest landscape in Central Amazonia. A taxonomically and trophically diverse assemblage of 993 species of beetles was sampled from 920 m2 of leaf litter at 46 sites varying in distance from forest edge and fragment area. Beetle density increased significantly towards the forest edge and showed non-linear changes with fragment area, due to the influx of numerous disturbed-area species into 10 ha and 1 ha fragments. There was a marked change in species composition with both decreasing distance from forest edge and decreasing fragment area, but surprisingly this change in composition was not accompanied by a change in species richness. Rarefied species richness did not vary significantly across any of the sites, indicating that local extinctions of deep forest species were balanced by equivalent colonization rates of disturbed-area species. The change in species composition with fragmentation was non-random across trophic groups. Proportions of predator species and xylophage species changed significantly with distance from forest edge, but no area-dependent changes in proportions of species in trophic groups were observed. Trophic structure was also analysed with respect to proportions of abundance in six trophic groups. Proportions of abundance of all trophic groups except xylomycetophages changed markedly with respect to both distance from forest edge and fragment area. Local extinction probabilities calculated for individual beetle species supported theoretical predictions of the differential susceptibility of higher trophic levels to extinction, and of changes in trophic structure following forest fragmentation. To reduce random effects due to sampling error, only abundant species (n ≥ 46) were analysed for extinction probabilities, as defined by absence from samples. Of these common species, 27% had significantly higher probabilities of local extinction following fragmentation. The majority of these species were predators; 42% of all abundant predator species were significantly more likely to be absent from samples in forest fragments than in undisturbed forest. These figures are regarded as minimum estimates for the entire beetle assemblage because rarer species will inevitably have higher extinction probabilities. Absolute loss of biodiversity will affect ecosystem process rates, but the differential loss of species from trophic groups will have an even greater destabilizing effect on food web structure and ecosystem function.