The affected sib-pair method has been used to investigate the mode of inheritance, and to estimate the 'disease' allele frequency, for a number of HLA-associated diseases. One of the assumptions of the original sib-pair method is that the disease confers no selective disadvantage on affected individuals. This is obviously not the situation for most diseases. We have determined the expected HLA haplotype-sharing distribution among affected sib-pairs when selection against individuals with the disease is taken into account. We have shown that if the mode of inheritance of the selectively disadvantageous disease is recessive or additive, the original affected sib-pair analysis, ignoring selection, still estimates the true mode of inheritance, but usually yields an underestimate of the 'disease' allele frequency. For intermediate and dominant models of disease predisposition, both the estimates of the degree of penetrance of the 'disease' genotypes, and the 'disease' allele frequency, are altered if selection is ignored in the analysis. Similarly, allowing for recombination between the 'disease' locus and the HLA region does not affect the determination of the mode of inheritance of the disease if it is recessive or additive; in other cases, however, the estimate of the mode of inheritance is affected. The 'disease' allele frequency is overestimated when nonzero recombination is ignored for all the modes of inheritance that have been studied.