For marine reserves to function as effective harvest refuges for exploited species, the reserve must protect a substantial proportion of the population for an indefinite period of time. Because most marine reserves are space-limited, the buildup and equilibrium population sizes of mobile species will be influenced by the size and boundary conditions of the refuge. A logistic rate model was used to predict equilibrium population sizes in a marine harvest refuge, based on species-specific dispersal dynamics and the spatial configuration of the refuge. The model parameters were derived for Caribbean spiny lobsters and queen conch in an isolated marine reserve at Glover’s Reef, Belize, and were compared to observed population change over a 5-yr period. Spiny lobsters and queen conch, the two most heavily exploited species in the Caribbean, differ in larval recruitment rates (immigration) and mobility of adults (emigration). The expected increase in the population size of spiny lobsters in this refuge was 250% and queen conch was 420% over that of the initial fished population. The observed densities of lobsters and conch in the refuge approached the predicted estimates within three years. To further explore the impact of alternative spatial configurations on refuge populations, the model was run on the same populations in two hypothetical refuges. In a refuge of the same area but 50% less absorbing boundary (adjacent to intensively fished areas), the spiny lobster population was expected to be 30% larger than the equilibrium population size in the original refuge, whereas the queen conch population was not expected to change from that in the original refuge. In a refuge that was 50% larger and with 50% less absorbing boundary, the spiny lobster population was expected to increase 110% and the queen conch population was expected to increase 50% over the equilibrium population size in the original refuge. Relatively minor changes in refuge area and boundary conditions may thus result in major population-level responses by exploited species, depending on dispersal dynamics and habitat availability. This simple model may be applicable for rapid assessment of the potential efficacy of proposed harvest refuges.