In this paper, an approach for the classification of dynamic models of diabetes mellitus is presented. The parameter vector of a personalized patient model, which has been identified e.g. by parameter estimation, is used as a classification feature. Principle component analysis and a support vector machine are used to reduce the feature space and to find a suitable classifier. The data covers type 1, type 2, and non-diabetic virtual subjects. Classification results show a good distinguishability between the classes, whereby the method may serve as a supplement in the area of model-driven diabetes management.