1. Risk effects of predators can profoundly affect community dynamics, but the nature of these effects is context dependent. 2. Although context dependence has hindered the development of a general framework for predicting the nature and extent of risk effects, recent studies suggest that such a framework is attainable if the factors that shape anti-predator behaviour, and its effectiveness, in natural communities are well understood. 3. One of these factors, the interaction of prey escape tactics and landscape features, has been largely overlooked. 4. We tested whether this interaction gives rise to interspecific variation in habitat-use patterns of sympatric large marine vertebrates at risk of tiger shark (Galeocerdo cuvier Peron and LeSueur, 1822) predation. Specifically, we tested the a priori hypothesis that pied cormorants (Phalacrocorax varius Gmelin, 1789) would modify their use of shallow seagrass habitats in a manner opposite to that of previously studied dolphins (Tursiops aduncus Ehrenberg, 1833), dugongs (Dugong dugon Müller, 1776), and green turtles (Chelonia mydas Linnaeus, 1758) because, unlike these species, the effectiveness of cormorant escape behaviour does not vary spatially. 5. As predicted, cormorants used interior and edge portions of banks proportional to the abundance of their potential prey when sharks were absent but shifted to interior portions of banks to minimize encounters with tiger sharks as predation risk increased. Other shark prey, however, shift to edge microhabitats when shark densities increase to take advantage of easier escape despite higher encounter rates with sharks. 6. The interaction of landscape features and escape ability likely is important in diverse communities. 7. When escape probabilities are high in habitats with high predator density, risk effects of predators can reverse the direction of commonly assumed indirect effects of top predators. 8. The interaction between landscape features and prey escape tactics can result in a single predator species having differential effects on their sympatric prey that could cascade through ecosystems and should be incorporated into a general framework for context dependence of risk effects.