Among Pima Indians with Type 2 (non-insulin-dependent) diabetes mellitus the relationships between glycated haemoglobin (HbA1), fasting or 2-h post-load plasma glucose and diabetic retinopathy were examined by cross-sectional and prospective analyses, and the strengths of the associations were directly compared by receiver operating characteristic analysis. In the cross-sectional analysis, HbA1, fasting and 2-h plasma glucose were each significantly related to retinopathy among 789 diabetic subjects by separate logistic models. In a stepwise multiple logistic model in which HbA1, fasting and 2-h plasma glucose were included, HbA1 was selected as having the strongest association with retinopathy and neither fasting nor 2-h plasma glucose contributed significantly to the model once HbA1 was entered. Similarly, in the prospective analysis, HbA1, fasting and 2-h plasma glucose all predicted retinopathy in 227 diabetic subjects by separate proportional-hazards models. In a stepwise proportional-hazards model with HbA1, fasting and 2-h plasma glucose available to the model, HbA1 was again selected as having the strongest association with the incidence of retinopathy, and neither fasting nor 2-h plasma glucose significantly added to the prediction of retinopathy. A receiver operating characteristic analysis was used to determine if HbA1 was statistically significantly better than fasting or 2-h plasma glucose in assessing the risk for retinopathy. In neither the cross-sectional nor the prospective data did the area under the receiver operating characteristic curve for HbA1 differ significantly from that for fasting or 2-h plasma glucose (p > 0.05 for each).(ABSTRACT TRUNCATED AT 250 WORDS)