By reducing genetically effective population size and gene flow, self-fertilization should lead to strong spatial genetic structure (SGS). Although the short-lived plant Aquilegia canadensis produces large, complex, nectar-rich flowers, 75% of seed, on average, are self-fertilized. Previous experimental results are consistent with the fine-scale SGS expected in selfing populations. In contrast, key floral traits show no evidence of SGS, despite a significant genetic basis to phenotypic variation within populations. In this study, we attempt to resolve these contradictory results by hierarchically sampling plants from two plots nested within each of seven rock outcrops distributed over several km, and comparing the spatial pattern of phenotypic variation in four floral traits with neutral genetic variation at 10 microsatellite loci. For both floral and microsatellite variation, we detected only weak hierarchical structuring and no isolation by distance. The spatial pattern of variation in floral traits was on par with microsatellite polymorphisms. These results suggest regular long-distance gene flow via pollen. At much finer spatial scales within plots, estimates of relatedness were higher (albeit very low) between nearest neighbours than random plants, and declined with increasing distance between neighbours, consistent with highly localized seed dispersal. High selfing should yield SGS, but strong inbreeding depression in A. canadensis likely erodes SGS so that reproductive plants exhibit weak structure typical of outcrossers, especially given that outcrossing and consequent gene flow in this species are mediated by strong-flying hummingbirds and bumble bees.