Premating isolation between incipient species is rarely studied in nature, even though mating tests in captivity may give an inaccurate picture of natural hybridization. We studied premating barriers between the warningly colored butterflies Heliconius erato and H. himera (Lepidoptera) in a narrow contact zone in Ecuador, where hybrids are found at low frequency. Eggs obtained from wild-mated females, supplemented with eggs and young larvae collected from the wild, were reared to adulthood. Adult color patterns of these progeny were then used to infer how their parents must have mated. Likelihood was used to estimate both the frequencies of potential parental genotypes from adult phenotypes collected in the wild, and the degree of assortative mating from the inferred parents. The frequencies of parental genotypes varied across the hybrid zone, but our statistical method allowed estimates of hybrid deficit and assortative mating to be integrated across all sites sampled. The best estimate of the frequency of F1 and backcross hybrid adults in the center of the hybrid zone was 10%, with support limits (7.1%, 13.0%; support limits are asymptotically equivalent to 95% confidence limits). Mating was highly assortative: in the center of the hybrid zone the cross-mating probability between H. erato and H. himera was only 5% (0.3%, 21.4%). Wild hybrids themselves mated with both pure forms, and the probabilities that they mated in any direction were not significantly lower than those among conspecifics. These results are consistent with earlier laboratory studies on mate choice, and suggest that selection against hybrids must be strong to prevent formation of a hybrid swarm. Unfortunately, the wide support limits on mating behavior precluded a measure of the strength of selection from these data alone. Our statistical approach provides a useful general method for estimating mate choice in the wild.