Objective: Practical use of the glycaemic index (GI), as recommended by the FAO/WHO, requires an evaluation of the recommended method. Our purpose was to determine the magnitude and sources of variation of the GI values obtained by experienced investigators in different international centres. Design: GI values of four centrally provided foods (instant potato, rice, spaghetti and barley) and locally obtained white bread were determined in 8-12 subjects in each of seven centres using the method recommended by FAO/WHO. Data analysis was performed centrally. Setting: University departments of nutrition. Healthy subjects (28 male, 40 female) were studied. Results: The GI values of the five foods did not vary significantly in different centres nor was there a significant centre´food interaction. Within-subject variation from two centres using venous blood was twice that from five centres using capillary blood. The s.d. of centre mean GI values was reduced from 10.6 (range 6.8-12.8) to 9.0 (range 4.8-12.6) by excluding venous blood data. GI values were not significantly related to differences in method of glucose measurement or subject characteristics (age, sex, BMI, ethnicity or absolute glycaemic response). GI values for locally obtained bread were no more variable than those for centrally provided foods. Conclusions: The GI values of foods are more precisely determined using capillary than venous blood sampling, with mean between-laboratory s.d. of approximately 9.0. Finding ways to reduce within-subject variation of glycaemic responses may be the most effective strategy to improve the precision of measurement of GI values.