Evidence for bimodality in the distribution of two hour post oral glucose challenge plasma glucose concentrations has come previously primarily from native American and Pacific Island populations having high non-insulin dependent diabetes mellitus (NIDDM) prevalence. Because the National Diabetes Data Group (NDDG) criteria for diagnosing NIDDM rely in part upon the assumption of bimodality, it is important to determine the generality of this phenomenon. We looked for bimodality among Mexican Americans in San Antonio, a population having greater than 50% Caucasian admixture. By fitting both a single normal distribution model and a mixture model of two normal distributions, for each age decade, we found that the mixture model was preferred to the single normal model (p less than 0.001) and that this model fit the data well. The proportion in the upper component (hyperglycemics) increased with each successive age decade. The minimum misclassification cutpoints decreased with age, but all were higher than the 200 mg/dl cutpoint recommended by the NDDG. Use of the NDDG cutpoint, however, improved sensitivity with only a minimal deterioration of specificity. Our findings further generalize the bimodality phenomenon and support the NDDG criteria.