To determine whether the diabetic exchange lists or the glycemic index of foods better predicts postprandial responses to carbohydrate-containing foods eaten as part of a mixed meal, three test meals were developed and fed to 12 subjects with non-insulin-dependent diabetes mellitus (NIDDM) and 13 healthy subjects. Each test meal contained exactly the same exchanges (1 milk, 4 starch, 2 fruit, 2 meat, 3 fat, 1 vegetable). In one meal, foods of high glycemic index (GI) were used, in a second meal, foods of intermediate GI were used, and in a third meal foods of low GI were used. The total GIs of the meals were: high, 184, intermediate, 131; and low, 107, thus predicting responses to intermediate and low GI, which were 71 and 58%, respectively, of the responses to high GI. Although some of the observed differences in the glycemic responses to the test meals were statistically significant, primarily in healthy subjects, the differences were usually much less than predicted by the GIs of the meals. In NIDDM subjects, peak postprandial plasma glucose, plasma glucose area, plasma glucose area increment, and mean plasma glucose responses after intermediate and low GI were > 90% of the corresponding responses to high GI. In healthy subjects, only the plasma glucose area increment after the low-GI meal was close to the predicted response. High GI produced significantly greater insulin responses than low GI in healthy subjects. We conclude that the diabetic exchange lists more accurately predict postprandial responses to carbohydrate-containing foods eaten as part of a mixed meal than does the GI of foods. Moreover, our data suggest that the responses of healthy subjects to test meals do not accurately predict the responses of NIDDM subjects.