The applicability of the glycaemic index (GI) in the context of mixed meals and diets is still debatable. The objective of the present study was to investigate the predictability of measured GI in composite breakfast meals when calculated from table values, and to develop prediction equations using meal components. Furthermore, we aimed to study the relationship between GI and insulinaemic index (II). The study was a randomised cross-over meal test including twenty-eight healthy young men. Thirteen breakfast meals and a reference meal were tested. All meals contained 50 g available carbohydrate, but differed considerably in energy and macronutrient composition. Venous blood was sampled for 2 h and analysed for glucose and insulin. Prediction equations were made by regression analysis. No association was found between predicted and measured GI. The meal content of energy and fat was inversely associated with GI (R2 0·93 and 0·88, respectively; P<0·001). Carbohydrate content (expressed as percentage of energy) was positively related to GI (R2 0·80; P<0·001). Using multivariate analysis the GI of meals was best predicted by fat and protein contents (R2 0·93; P<0·001). There was no association between GI and II. In conclusion, the present results show that the GI of mixed meals calculated by table values does not predict the measured GI and furthermore that carbohydrates do not play the most important role for GI in mixed breakfast meals. Our prediction models show that the GI of mixed meals is more strongly correlated either with fat and protein content, or with energy content, than with carbohydrate content alone. Furthermore, GI was not correlated with II.