Previously published Type I (insulin-dependent) diabetes mellitus incidence in 0 to 14-year-old children from 10 countries or areas was compared with the national annual cow milk protein consumption. Countries which were selected for study had appropriate milk protein polymorphism studies, herd breed composition information and low dairy imports from other countries. Total protein consumption did not correlate with diabetes incidence (r = + 0.402), but consumption of the β-casein A¹ variant did (r = + 0.726). Even more pronounced was the relation between β-casein (A¹ + B) consumption and diabetes (r = + 0.982). These latter two cow caseins yield a bioactive peptide β-casomorphin-7 after in vitro digestion with intestinal enzymes whereas the common A² variant or the corresponding human or goat caseins do not. β-casomorphin-7 has opioid properties including immunosuppression, which could account for the specificity of the relation between the consumption of some but not all β-casein variants and diabetes incidence. [Diabetologia (1999) 42: 292–296]