The spatial scale of larval dispersal is a key predictor of marine metapopulation dynamics and an important factor in the design of reserve networks. Over the past 15 yr, studies of larval dispersal in coral reef fishes have generated accumulating evidence of consistently high levels of self-recruitment and local retention at various spatial scales. These findings have, to a certain degree, created a paradigm shift toward the perception that large fractions of locally produced recruitment may be the rule rather than the exception. Here we examined the degree of localized settlement in an anemonefish, Amphiprion bicinctus, at a solitary coral reef in the central Red Sea by integrating estimates of self-recruitment obtained from genetic parentage analysis with predictions of local retention derived from a biophysical dispersal model parameterized with real-time physical forcing. Self-recruitment at the reef scale (c. 0.7 km2) was virtually absent during two consecutive January spawning events (1.4 % in 2012 and 0 % in 2013). Predicted levels of local retention at the reef scale varied temporally, but were comparatively low for both simulations (7 % in 2012 and 0 % in 2013). At the same time, the spatial scale of simulated dispersal was restricted to approximately 20 km from the source. Model predictions of reef-scale larval retention were highly dependent on biological parameters, underlining the need for further empirical validations of larval traits over a range of species. Overall, our findings present an urgent caution when assuming the potential for self-replenishment in small marine reserves.