The persistence of prey encountering intense predation varies by species, prey density, and habitat type; however, the collective impact of these factors has rarely been tested experimentally in natural marine systems. Using the thin-shelled clams Mya arenaria and Macoma balthica as prey, and the main epibenthic predator of whole adult clams, the blue crab Callinectes sapidus, we conducted a series of experiments in Chesapeake Bay tributaries that (1) links field abundance and distribution of bivalve prey species with habitat-specific mortality patterns; (2) represents the first comprehensive field test of species-specific, habitat-specific, and density-dependent mortality for subtidal, soft-bottom, deep-burrowing prey; and (3) thereby enables development of a conceptual model to be used as a heuristic tool linking predator-prey dynamics, habitat type, and evolutionary defense tactics for marine benthos. In 15 years of field monitoring, Mya was more common in sand than mud habitats, and Macoma was widely distributed and at higher densities than Mya in mud and sand. In field experiments, mortality of both Mya and Macoma was density dependent in those habitats where the clams are common. The blue crab population in the field exhibited a type III "guild functional response" on Mya in sand, and on Macoma in both mud and sand. Mortality was lower in sand than mud for Mya, and similar in mud and sand for Macoma, correlating with the high abundances of Mya in sand and Macoma in sand and mud. The persistence of large juvenile and adult bivalves when confronted with intense predation derived substantially from a low-density refuge from predation that varied in a species-specific manner with habitat type, demonstrating the species-specific importance of density and habitat to clam survival. We developed a conceptual model detailing the relative importance of behavior, morphology, habitat features, and the basic components of predator-prey interactions to the survival of bivalve molluscs. At one extreme are bivalve molluscs, such as oysters, that emphasize morphological refuges that increase the predator's handling time. At the other extreme are bivalves, such as Mya and Macoma, that reduce predator encounter rates. The model is intended to be used as a heuristic tool to develop testable hypotheses.