Many marine organisms can be transported hundreds of kilometers during their pelagic larval stage, yet little is known about spatial and temporal patterns of larval dispersal. Although traditional population-genetic tools can be applied to infer movement of larvae on an evolutionary time scale, large effective population sizes and high rates of gene-flow present serious challenges to documenting dispersal patterns over shorter, ecologically-relevant, time scales. Here, we address these challenges by combining direct parentage analysis and indirect genetic analyses over a four-year period to document spatial and temporal patterns of larval dispersal in a common coral-reef fish: the bicolor damselfish (Stegastes partitus). At four island locations surrounding Exuma Sound, Bahamas, including a long-established marine reserve, we collected 3,278 individuals and genotyped them at 10 microsatellite loci. Using Bayesian parentage analysis, we identified eight parent-offspring pairs, thereby directly documenting dispersal distances ranging from 0 km (i.e., self-recruitment) to 129 km (i.e., larval connectivity). Despite documenting substantial dispersal and gene-flow between islands, we observed more self-recruitment events than expected if the larvae were drawn from a common, well-mixed pool (i.e., a completely open population). Additionally, we detected both spatial and temporal variation in signatures of sweepstakes and Wahlund effects. The high variance in reproductive success (i.e., “sweepstakes”) we observed may be influenced by seasonal mesoscale gyres present in the Exuma Sound, which play a prominent role in shaping local oceanographic patterns. This study documents the complex nature of larval dispersal in a coral-reef fish, and highlights the importance of sampling multiple cohorts and coupling both direct and indirect genetic methods in order disentangle patterns of dispersal, gene-flow, and variable reproductive success.