Principles of probability survey design were applied to guide large-scale sampling of populations of stony corals and associated benthic taxa in the Florida Keys coral reef ecosystem. The survey employed a two-stage stratified random sampling design that partitioned the 251-km(2) domain by reef habitat types, geographic regions, and management zones. Estimates of the coefficient of variation (ratio of standard error to the mean) for stony coral population density and abundance ranged from 7% to 12% for four of six principal species. These levels of survey precision are among the highest reported for comparable surveys of marine species. Relatively precise estimates were also obtained for octocoral density, sponge frequency of occurrence, and benthic cover of algae and invertebrates. Probabilistic survey design techniques provided a robust framework for estimating population-level metrics and optimizing sampling efficiency.