Studies of natural hybridization have suggested that it may be a creative stimulus for adaptive evolution and speciation. An important step in this process is the establishment of fit recombinant genotypes that are buffered from subsequent recombination with unlike genotypes. We used molecular markers and a two-generation sampling strategy to infer the extent of recombination in a Louisiana iris hybrid zone consisting predominantly of Iris fulva– type floral phenotypes. Genotypic diversity was fairly high, indicating that sexual reproduction is frequent relative to clonal reproduction. However, we observed strong spatial genetic structure even after controlling for clonality, which implies a low level of pollen and seed dispersal. We therefore used cluster analysis to explore the hypothesis that the fulva-type hybrids are an admixture of groups between which there has been limited recombination. Our results indicate that several such groups are present in the population and are strongly localized spatially. This spatial pattern is not attributable strictly to a lack of mating opportunities between dissimilar genotypes for two reasons: (1) relatedness of flowering pairs was uncorrelated with the degree of overlap in flowering, and (2) paternity analysis shows that pollen movement among the outcross fraction occurred over large distances, with roughly half of all paternity attributed to pollen flow from outside the population. We also found evidence of strong inbreeding depression, indicated by contrasting estimates of the rate of self-fertilization and the average inbreeding coefficient of fulva-type hybrids. We conclude that groups of similar hybrid genotypes can be buffered from recombination at small spatial scales relative to pollen flow, and selection against certain recombinant genotypes may be as important as or more important than clonal reproduction and inbreeding.