The extent to which random plasma glucose levels underestimate the true prevalence of diabetes has been determined in Micronesian Nauruans. In 337 individuals who were screened on the basis of their random plasma glucose levels, the age-standardised prevalence based on a cut-off of 11.1 mmol/l underestimated the population prevalence based on a complete oral glucose tolerance test by 42% in males and 63% in females. At a cut-off level of 7.8 mmol/l the true age-standardised prevalence was underestimated by 16 and 38%, in males and females, respectively. The use of random plasma glucose concentrations to determine the prevalence of diabetes, as currently defined, seems inappropriate. Performing oral glucose tolerance tests on smaller representative population samples should provide more accurate data at less expense than through large-scale screening utilizing random glucose levels.