IDDM2-encoded predisposition to type 1 diabetes has recently been mapped to the minisatellite or variable number of tandem repeat (VNTR) locus upstream of the insulin and insulin-like growth factor II genes on human chromosome 11p15.5. In a UK case-control study (n = 228 sporadic diabetics; n = 441 healthy controls), we show here that the genotype homozygous for VNTR class I alleles is predisposing to disease (RR = 2.68), and VNTR class III alleles are dominantly protective (RR = 0.37). In 722 diabetic families from the UK (n = 356), USA (n = 173), Denmark (n = 55) and Sardinia (n = 138), we have analysed the transmission of class I alleles to diabetic offspring from class I/III heterozygous parents. We confirm that in families from the USA, class I alleles are transmitted preferentially from fathers. However, in family data sets from the UK, Denmark and Sardinia, the reverse is true and maternal transmission is stronger. Furthermore, in the UK family data set, the difference between maternal and paternal transmissions is significant (P < 0.05). It is therefore unlikely that 'maternal imprinting' alone explains the parent-of-origin effects in IDDM2-encoded predisposition to type 1 diabetes, at least not in the UK. There is a relationship between VNTR class (allele length) and insulin gene expression, though some results from different studies are conflicting. In the human adult cadaveric pancreas, we confirm our preliminary results that class III alleles are associated with lower levels of insulin mRNA in vivo. Similar results have been obtained independently in human foetal pancreas samples. It is difficult to explain how these marginally lower levels of insulin expression could account for the observed VNTR class III-encoded protective effect. Perhaps the site of action of IDDM2, mediated by VNTR allelic variation, is not the pancreas but some other organ such as the thymus.