Aim Working within a system of high structural contrast between fragments and the surrounding matrix, we assessed patterns of species loss and changes in species composition of phyllostomid bats on artificial land-bridge islands relative to mainland assemblages, and evaluated the responses of bats to forest edges. We further examined the relative influence of local-scale characteristics (e.g. vegetation structure, island area) versus landscape attributes (e.g. forest cover, patch density) and the importance of spatial scale in determining phyllostomid species richness and composition on islands.Location Islands in Gatiin Lake and adjacent mainland peninsulas in the Barro Colorado Nature Monument, Panama.Methods Bats were sampled over a 2-year period on 11 islands as well as at forest-edge and interior sites on adjacent mainland, resulting in > 8400 captures.Results The islands harboured a less diverse and structurally simplified phyllostomid bat fauna. Islands far from the mainland were especially speciespoor. This decline in species richness was associated with compositional shifts towards assemblages strongly dominated by frugivores with good dispersal abilities. Members of other ensembles, most importantly gleaning animalivores, were much less common or absent. Although overall species composition was not significantly altered, species richness at continuous forest-edge sites was significantly lower compared with that at interior sites. Distance from the mainland and amount of forest cover in the landscape were the best predictors of species richness and assemblage composition. Responses were scale-dependent. At the local scale, species richness was independent of island area but was correlated positively with distance from the mainland. In contrast, area effects became more important at larger spatial scales, suggesting that many species use multiple fragments.Main conclusions Our results underline the conservation value of small habitat remnants, which, even when embedded in a hostile matrix, can support a relatively diverse bat fauna, provided that there is a low degree of patch isolation and spatial proximity to larger tracts of continuous forest. Although the results at the assemblage level were inconclusive, we demonstrate that certain bat species and ensembles, particularly gleaning animalivores, exhibit high edge-sensitivity. Our results point to habitat loss rather than changes in landscape configuration as the main process after isolation underlying phyllostomid bat responses, suggesting that conservation efforts should focus on habitat preservation instead of trying to minimize fragmentation per se at the expense of habitat amount.