The level of site fidelity in birds is often characterized as "high" on the basis of rates of return or homing from mark–recapture data. For species that exhibit site fidelity, subsequent biological assumptions have included population structure, demographic independence, and that the extirpation of a site-faithful group might be irreversible because of low immigration. Yet several genetic studies have observed patterns of population differentiation that are incongruous with strong site fidelity, which suggests recent isolation, gene flow, or both. Using a 13-year live-recapture and dead-recovery data set, as well as nuclear and mitochondrial DNA collected across the range of the Hooded Merganser (Lophodytes cucullatus), an obligate cavity-nester endemic to North America, we found evidence that gene flow persists across portions of the species' range even though the probability of female breeding-site fidelity is high (0.92; 95% confidence interval [CI]: 0.64–0.98) and disjunct breeding ranges of this species have been isolated for ≥10,000 years. By combining inferences from genetic, band-recovery, mark–recapture, and comparative data from another cavity-nesting species of waterfowl, we conclude that a high level of site fidelity should not be considered a universal proxy for population structure and demographic independence. Our results also suggest that an accurate assessment of site fidelity—and its implications for population dynamics and delineation—requires cross-species comparisons and multiple data types, such as mark–recapture and genetic information, to best infer patterns across a range of geographic and temporal scales.