Aims/hypothesis. The aim of this study was to assess the effect of ethnicity on the association between age and body mass index as well as the prevalence of diabetes. Methods. We selected population-based studies carried out after 1980 in the DECODE/A studies representing different ethnic groups: 11 European, 1 Maltese, 3 Indian, 2 Chinese and 3 Japanese surveys. The total numbers of subjects were 14,240 men and 15,129 women who were 30 to 89 years of age. Diabetes was diagnosed according to the 1999 World Health Organization criteria based on a standard 75 g OGTT. Sex-specific prevalence of diabetes by age and BMI was stratified by ethnic group, in particular the interaction of ethnicity on the associations between age/BMI and the prevalence of diabetes. Results. The prevalence of diabetes was higher in studies from India and Malta compared to Japan, China, and the rest of Europe. The association between BMI and diabetes, adjusted for age, showed noticeable differences between the ethnic groups with an increase in prevalence starting at a BMI between 15 and 20 kg/m2 in the Maltese and Indian populations compared to 25 kg/m2 in Europeans. Conclusion/interpretation. The effect of BMI on the age-adjusted prevalence of Type 2 diabetes was modified by ethnicity with considerably lower thresholds in Indian and Maltese subjects compared to those from the rest of Europe. This difference should be reflected in national and international recommendations regarding "optimal" BMI.