The aim of the study was to compare BMI with waist circumference (WC), waist-to-hip ratio (WHR), and waist-to-stature ratio (WSR) as a predictor of diabetes incidence. A total of 1,841 men and 2,104 women of Mauritian Indian and Mauritian Creole ethnicity, aged 25-74 years, free of diabetes, hypertension, cardiovascular disease, and gout were seen at baseline in 1987 or 1992, and follow-up in 1992 and/or 1998. At all time points, participants underwent a 2 h 75 g oral glucose tolerance test. Hazard ratios for diabetes incidence were estimated applying an interval-censored survival analysis using age as timescale. Six hundred and twenty-eight individuals developed diabetes during the follow-up period. Multivariable adjusted hazard ratios for diabetes incidence corresponding to a 1 s.d. increase in baseline BMI, WC, WHR, and WSR for Mauritian Indians were 1.49 (1.31-1.71), 1.58 (1.38-1.81), 1.54 (1.37-1.72), and 1.61 (1.41-1.84) in men and 1.33 (1.17-1.51), 1.35 (1.19-1.53), 1.39 (1.24-1.55), and 1.38 (1.21-1.57) in women, respectively; and for Mauritian Creoles they were 1.86 (1.51-2.30), 2.07 (1.68-2.56), 1.92 (1.62-2.26), and 2.17 (1.76-2.69) in men and 1.29 (1.06-1.55), 1.27 (1.04-1.55), 1.24 (1.04-1.48), and 1.27 (1.04-1.55) in women. Paired homogeneity tests showed that there was no difference between BMI and each of the central obesity indicators (all P > 0.05). The relation of BMI with the development of diabetes was as strong as that for indicators of central obesity in this study population.