In a substructured population the overall heterozygote deficiency can be predicted from the number of subpopulations (s), their time of divergence (t), and the nature of the mutations. At present the true mutational mechanisms at the hypervariable DNA loci are not known. However, the two existing mutation models (the infinite allele model (IAM) and the stepwise mutation model (SMM)) provide some guides to predictions from which the possible effect of population substructuring may be evaluated, assuming that the subpopulations do not exchange any genes among them during evolution. The theory predicts that the loci with larger mutation rate, and consequently showing greater heterozygosity within subpopulations, should exhibit a smaller proportional heterozygote deficiency (GST) and, hence, the effects of population substructuring should be minimal at the hypervariable DNA loci (an order of magnitude smaller than that at the blood group and protein loci). Applications of this theory to data on six Variable Number of Tandem Repeat (VNTR) loci and five short tandem repeat (STR) loci in the major cosmopolitan populations of the USA show that while the VNTR loci often exhibit a large significant heterozygote deficiency, the STR loci do not show a similar tendency. This discordant finding may be ascribed to the limitations, coalescence and nondetectability of alleles associated with the restriction fragment length polymorphism (RFLP) analysis through which the VNTR loci are scored. Such limitations do not apply to the polymerase chain reaction (PCR) method, through which the STR loci are scored. The implications of these results are discussed in the context of the forensic use of DNA typing data.