Unexpected plasma glucose responses to different mixed meals fed to normal and diabetic volunteers have recently been reported. We have therefore examined in normal volunteers the effect of mixing carbohydrate foods of different glycemic indices (GIs) without the addition of fat and protein. The observed GI of the mixed meal was within 2% of the expected value. In studies in the literature where fat and protein were added to mixed meals, the observed blood glucose responses also related significantly to the meal GIs calculated from the individual foods. Addition of fat and protein in the quantities used did not obscure this relationship. Studies to determine sources of error in comparing glycemic responses showed that type II diabetic patients displayed the least within-individual variation, and type I diabetic patients the most. Expression of results as the GI rather than as absolute glycemic response areas reduced by 50% the between-subject variation. The mean GI values of rice tested in type I and type II patients were similar (82 ± 22 compared with 74 ± 19) and the reproducibility 22 mo later in the same group of subjects was excellent (81 ± 15 compared with 83 ± 15). However, the lack of precise GI values for all foods fed in the test meals indicates a need for GI values to be derived for a wider range of individual foodstuffs. The GI approach to classifying foods according to physiologic effect may play a useful role in planning meals and diets in which specific blood glucose profiles are required.