Metapopulation dynamics are increasingly invoked in management and conservation of endangered species. In this context, asymmetrical gene flow patterns can be density dependent, with migration occurring mainly from larger into smaller populations, which may depend on it for their persistence. Using genetic markers, such patterns have recently been documented for various organisms including salmonids, suggesting this may be a more general pattern. However, metapopulation theory does not restrict gene flow asymmetry to ‘source-sink’ structures, nor need these patterns be constant over longer evolutionary timescales. In anadromous salmonids, gene flow can be expected to be shaped by various selective pressures underlying homing and dispersal (‘straying’) behaviours. The relative importance of these selective forces will vary spatially and for populations of different census size. Furthermore, the consequences of life-history variation among populations for dispersal and hence gene flow remain poorly quantified. We examine population structure and connectivity in Atlantic salmon (Salmo salar L.) from Newfoundland and Labrador, a region where populations of this species are relatively pristine. Using genetic variation at 13 microsatellite loci from samples (N = 1346) collected from a total of 20 rivers, we examine connectivity at several regional and temporal scales and test the hypothesis that the predominant direction of gene flow is from large into small populations. We reject this hypothesis and find that the directionality of migration is affected by the temporal scale over which gene flow is assessed. Whereas large populations tend to function as sources of dispersal over contemporary timescales, such patterns are often changed and even reversed over evolutionary, coalescent-derived timescales. These patterns of population structure furthermore vary between different regions and are compatible with demographic and life-history attributes. We find no evidence for sex-biased dispersal underlying gene flow asymmetry. Our findings caution against generalizations concerning the directionality of gene flow in Atlantic salmon and emphasize the need for detailed regional study, if such information is to be meaningfully applied in conservation and management of salmonids.