Fifty nine insulin dependant diabetics were hospitalised for a trial withdrawal of insulin: 17 patients rapidly showed signs of lack of insulin, 18 did not develop cetoacidosis but could not be stabilised on diet and oral hypoglycemic agents, 24 were stabilised without insulin. A statistical study (multifactorial analysis of correlations, plotting of ROC graphs) validated the classification of these diabetics into 3 groups. It also showed that in patients with hypoglycemia, the values of C-protein, and after intravenous injection of tolbutamide, were good predictive factors for insulin-dependance: all patients with basal C-protein less than 1,9 ng/ml could not be stabilised without insulin; when the basal C-protein greater than or equal to 1,9 ng/ml and the amplitude of response at the 5th min was greater than or equal to 0,4 ng/ml, the diabetes could be stabilised by diet and oral hypoglycemic agents in 90 p. 100 of cases. This institutes an easy, reliable and economic method of detecting abusive insulin therapy.