Objectives - To propose clinical and bio-clinical scores predictive of diabetes in the French population. Method - The study included 1,863 men and 1,954 women (D.E.S.I.R. cohort), aged 30–65 years at enrollment and followed for nine years. New-onset diabetes was identified by treatment or a fasting blood glucose level of 7.0 mmol/L at one of the three triennial examinations. The most predictive variables were identified using logistic regression. Results - A total of 140 men and 63 women developed diabetes. The clinical variables predictive of diabetes onset were waist circumference and hypertension, as well as smoking for men and a family history of diabetes for women. The area under the ROC curve was 0.71 for men and 0.83 for women, significantly better performance than that of the Findrisc score, using fewer variables. The addition of biological variables (blood glucose, GGT, triglycerides) improved prediction. Conclusions - Adiposity is the most predictive clinical variable for diabetes at 9 years. By adding hypertension and smoking in men, and hypertension and a family history of diabetes in women, a highly practical clinical score is obtained for general practitioners to target preventive measures for diabetes. (R.A.)