The preservation of biodiversity requires an understanding of the maintenance of its components, including genetic diversity. Effective population size determines the amount of genetic variance maintained in populations, but its estimation can be complex, especially when populations are interconnected in a metapopulation. Theory predicts that the effective size of a metapopulation (meta-N(e)) can be decreased or increased by population subdivision, but little empirical work has evaluated these predictions. Here, we use neutral genetic markers and simulations to estimate the effective size of a putative metapopulation in Atlantic salmon (Salmo salar). For a weakly structured set of rivers, we find that meta-N(e) is similar to the sum of local deme sizes, whereas higher genetic differentiation among demes dramatically reduces meta-N(e) estimates. Interdemic demographic processes, such as asymmetrical gene flow, may explain this pattern. However, simulations also suggest that unrecognized population subdivision can also introduce downward bias into empirical estimation, emphasizing the importance of identifying the proper scale of distinct demographic and genetic processes. Under natural patterns of connectivity, evolutionary potential may generally be maintained at higher levels than the local population, with implications for conservation given ongoing species declines and habitat fragmentation.