The Great Barrier Reef (GBR) is a continental archipelagic system of 5000 reefs and shoals stretching >2000 km along the east Australia coast. The interconnectivity of these reefs should determine the choice of biological management units, which for most biota will reflect the dispersal of their eggs and/or larvae. A comparative approach using population genetics was used to ask whether the along—shore dispersal of coral reef fishes is influenced by the duration of this mobile phase. Seven species of coral reef fish, selected from three families to provide a range of taxonomic diversity and pelagic larval durations, were tested for genetic homogeneity between two regions of the GBR separated by 1000 km. A spectrum of potential dispersal capabilities was analyzed from that of Acanthochromis polyacanthus, a damselfish with brood care that uniquely lacks pelagic larvae, to that of Ctenochaetus striatus, a surgeonfish with large, specialized larvae that spend several months in the plankton. A total of 19 enzyme systems and general proteins were examined from multiple populations in each region to provide a base of 32 loci for these comparisons. With one exception, species sampled from different coral reefs within regions showed statistically significant heterogeneities across multiple loci, indicative of chaotic genetic patchiness among the samples. The exception was an anemonefish, Amphiprion melanopus, that had to be collected from large areas on each reef because of its low densities. The homogeneity of allele frequencies at local scales for this species suggests that the genetic patchiness observed in others may be a within—reef phenomenon that was manifested at the reef scale by our pseudoreplicated sampling strategy. After pooling local variability, all but two species showed significant regional differences. The exceptions were the pair (Ctenochaetus striatus, Pterocaesio chrysozona) with the longest larval durations. Acanthochromis polyacanthus showed increased variation at this larger scale, consistent with a major stock division between the two regions. The logarithm of genetic variation between northern and southern populations (measured by Weir and Cockerham's Fst was correlated with mean larval duration by an inverse linear relationship that explained 85% of the variance in the global data set. Comparison with an outgroup (Amphiprion melanopus from the Chesterfield Reefs, 1000 km east in the Coral Sea) confirmed the genetic cohesion of mainland populations for the species with shortest larval duration and shows that our empirical relationship applies only within the context of the highly connected GBR. On this basis, calculations of gene flow (Nem, the number of effective migrants per generation) between geographic regions predict panmixis for species with larval durations exceeding 1 mo. Many common species have shorter dispersal times, from which classical isolation—by—distance" models predict differentiation between northern and southern populations at genetic equilibrium. Given that modern populations on the GBR are <10 000 yr old, however, there has not been sufficient time for such differences to evolve in situ and we consider alternative scenarios for the observed heterogeneities. Comparisons with invertebrate taxa sampled over the same spatial scales imply lower gene flows in fish despite longer pelagic durations. This suggests that fish larvae may use their greater mobility to retard, rather than enhance, dispersal due to hydrodynamic advection.