In order to investigate the performance of haplotype frequency estimation methods using unrelated individuals, we compared the results of three estimation methods with those from the haplotypes deduced from family pedigrees. To that end we used the HLA phenotypes of the parents of 1040 families as data for the estimation methods and the full pedigree information as data for the deductive method. We evaluated the results of the following estimation methods: the method using two by two tables described by Mattiuz et al., the maximum likelihood method described by Yasuda and Tsuji and a crude method that uses the information on homozygosity in the phenotypes. All estimation methods generate reliable haplotype frequencies for the more frequent haplotypes, but are unreliable for the less frequent haplotypes. The maximum likelihood estimation method shows the best overall correlation with the results of the deductive method.