The admixture test (A-test) is a popular method for the analysis of linkage data when locus heterogeneity is suspected. It can be applied on pairwise linkage data, multipoint data and even for the simultaneous analysis of data from multiple dispersed candidate regions. However, very little is known about the conditions for the use of the method under these divergent circumstances. By performing analytical evaluations, we demonstrate that the A-test is inconsistent if there is a relationship between the phenotype and the probability of being linked. Biased estimates of the recombination fraction (Θ) and the proportion of linked families (α) may occur if the actual frequency of linked families is not identical among small and large families. We conclude that the A-test should be used with caution if the phenotype and the probability of developing the phenotype at a certain age cannot be shown to be equal for family members of linked and unlinked families. If dissimilarities in family size cannot be ruled out, the extent of bias should be considered and size specific α-values should be used in risk calculations.