Prediction of the disease status is one of the most important objectives of genetic studies. To select the genes with strong evidence of the association with type 2 diabetes mellitus, we validated the associations of the seven candidate loci extracted in our earlier study by genotyping the samples in two independent sample panels. However, except for KCNQ1, the association of none of the remaining seven loci was replicated. We then selected 11 genes, KCNQ1, TCF7L2, CDKAL1, CDKN2A/B, IGF2BP2, SLC30A8, HHEX, GCKR, HNF1B, KCNJ11 and PPARG, whose associations with diabetes have already been reported and replicated either in the literature or in this study in the Japanese population. As no evidence of the gene–gene interaction for any pair of the 11 loci was shown, we constructed a prediction model for the disease using the logistic regression analysis by incorporating the number of the risk alleles for the 11 genes, as well as age, sex and body mass index as independent variables. Cumulative risk assessment showed that the addition of one risk allele resulted in an average increase in the odds for the disease of 1.29 (95% CI= 1.25–1.33, P= 5.4× 10− 53). The area under the receiver operating characteristic curve, an estimate of the power of the prediction model, was 0.72, thereby indicating that our prediction model for type 2 diabetes may not be so useful but has some value. Incorporation of data from additional risk loci is most likely to increase the predictive power.