Body mass index (BMI) is an established epidemiological predictor of coronary disease, diabetes and hypertension. In a previous study of 2560 healthy British Caucasoid males aged 50-61 years (Northwick Park Heart Study II; NPHSII), we showed that IGF2 ApaI AA homozygotes display a mean body weight 3.3 kg lower than GG homozygotes (P = 0.0002) independent of height. Two RFLPs in the insulin (INS) gene, +1127/PstI shown previously and -23/HphI in this study, both of which are in strong linkage disequilibrium with class I/III alleles of the INS 5' variable number tandem repeat (VNTR), are not associated with weight or BMI. The IGF2 ApaI polymorphism therefore appears to mark an effect independent of INS VNTR class I vs class III. We now show by regression that there is a positive correlation of BMI with INS VNTR class I allele size, with an average 0.33% (95% CI = 0.13%, 0.50%) increase in BMI per extra tandem repeat (P < 0.0001) representing variation of 4.8% over the allele size range. However, an alternative interpretation is of 'step' rather than 'slope', the small class I subclass allele group (mode 669 bp) being lighter than the large subclass group (mode 814 bp). This small effect would not be evident as an association between INS VNTR class I/I1 genotype and BMI. The IGF2 ApaI association and INS VNTR class I subclass regression association account for at least 1.1% of population BMI variance. Neither, both, or a third site may be aetiological.