OBJECTIVES: Modelling approaches are used to estimate future consequences and costs of diabetes and its complications, since few studies provide data over a sufficient duration or for all costs, effects and populations. The Diabetes Mellitus Model (DMM) predicts 10-year outcomes for patients with Type 1 (T1DM) and Type 2 (T2DM) diabetes based on published clinical trials; these predictions mirror clinical trial outcomes. We compared the DMM against real world, population-based epidemiological data from the Diabetes Audit and Research in Tayside Scotland (DARTS) database. METHODS: We studied two cohorts of 931 T1DM (46% male) and 12,907 T2DM (53% male) patients, mean (±SD) age at diagnosis 21 (±14) and 61 (±13) years, respectively. Mean A1c for DARTS patients, diagnosed from 1 January 1993 to 31 December 2000, was derived as a function of time since, and age at, diagnosis. Cohorts were simulated using the DMM, and resulting A1c values were compared with DARTS cohorts. Goodness of fit was evaluated by assessing bias, i.e. the underlying difference between DMM and DARTS, and stochastic variation. RESULTS: For patients with T1DM, changes in A1c over time were not predicted well by the DMM; many of the differences between DARTS and DMM predicted A1c values were greater than 0.5% and the maximum bias was 0.9%. For patients with T2DM, the changes were well predicted for all ages and age bands; maximum bias was 0.5%. CONCLUSION: The DMM is successful in predicting real world changes in A1c for T2DM; further work is needed to reproduce real world changes in A1c for T1DM.