Habitat fragmentation involves habitat loss concomitant with changes in spatial configuration, confounding mechanistic drivers of biodiversity change associated with habitat disturbance. Studies attempting to isolate the effects of altered habitat configuration on associated communities have reported variable results. This variability may be explained in part by the fragmentation threshold hypothesis, which predicts that the effects of habitat configuration may only manifest at low levels of remnant habitat area. To separate the effects of habitat area and configuration on biodiversity, we surveyed fish communities in seagrass landscapes spanning a range of total seagrass area (2-74% cover within 16 000-m landscapes) and spatial configurations (1-75 discrete patches). We also measured variation in fine-scale seagrass variables, which are known to affect faunal community composition and may covary with landscape-scale features. We found that species richness decreased and the community structure shifted with increasing patch number within the landscape, but only when seagrass area was low (<25% cover). This pattern was driven by an absence of epibenthic species in low-seagrass-area, highly patchy landscapes. Additional tests corroborated that low movement rates among patches may underlie loss of vulnerable taxa. Fine-scale seagrass biomass was generally unimportant in predicting fish community composition. As such, we present empirical support for the fragmentation threshold hypothesis and we suggest that poor matrix quality and low dispersal ability for sensitive taxa in our system may explain why our results support the hypothesis, while previous empirical work has largely failed to match predictions.