The aim is to analyse the association between social network and self-rated health in a longitudinal design for persons with type 2 diabetes, comparing them with persons without diabetes. METHODS: The analyses are based on data from the population-based 'KORA-A study' conducted in the region of Augsburg (Germany), with data from 1990/1995 (t (0)) and 1998 (t (1)), including 164 persons with type 2 diabetes and 207 persons without diabetes. The social network was assessed by the Berkman/Syme index. RESULTS: For the type 2 diabetes group, the multivariate analyses show that a high score of social network at t (0) is associated with good self-rated health at t (1), even if self-rated health at t (0) is included in the model (OR 2.69; 95 % CI: 1.21-5.98). For the non-diabetes group, no such association was found. CONCLUSIONS: The results point towards a 'buffer effect' of the social network, indicating that the positive effect on health can be seen mostly among those who are exposed to a high level of burden, in this case exposed to a chronic disease such as type 2 diabetes.