The fine-scale genetic structure of wild animal populations has rarely been analysed, yet is potentially important as a confounding factor in quantitative genetic and allelic association studies, as well as having implications for population dynamics, inbreeding and kin selection. In this study, we examined the extent to which the three spatial subunits, or hefts, of the Village Bay population of Soay sheep (Ovis aries) on St Kilda, Scotland, are genetically structured using data from 20 microsatellite and protein loci. Allele frequencies differed significantly among three hefts in all the study years we considered (1987–2000 inclusive). Small but significantly positive FST and negative FIS values were observed in most years, indicating that the hefts are genetically differentiated, and that within each heft there is more observed heterozygosity than would be expected if each were an isolated breeding population. Males showed less fidelity to their natal heft, and as a consequence higher levels of relatedness within hefts were observed among females than among males. There was a significant negative relationship between geographical proximity and relatedness in pairwise comparisons involving females, and on average pairs of females located within 50 m of each other were related at the equivalent level of second cousins. Structure is therefore largely driven by incomplete postnatal dispersal by females. Mating appears to be random with respect to the spatial-genetic substructure of the hefts, and therefore genetic structure does not contribute to the overall rate of inbreeding in the population. However, genetic substructure can lead to allelic associations and generate environmental effects within lineages that have the potential to confound heritability analyses and allelic association studies.