An epidemiologic study of end-stage diabetic nephropathy in France (Uremidiab) was performed, aiming to establish the prevalence of both types of diabetes in dialysis patients. Because discrimination between type I and type II diabetes remains mostly clinical, our aim was to evaluate what the most fitted clinical criteria were. We studied 494 hemodialyzed diabetic patients. A first classification (Cn) was offered by the nephrologist. Clinical data of 472 patients (22 patients of the 494 have been excluded) were then collected with a standardized questionnaire, allowing one diabetologist of us to establish the diagnosis of type of diabetes (classification Cd). Plasma C-peptide at this stage of the disease was expected to be very discriminative, measured in 88 patients and defined classification Ccp (< or = 0.6 ng/ml = "negative C-peptide" = type I, > 0.6 ng/ml = "positive C-peptide" = type II). Classification Cd observed 98 type I and 374 type II diabetes. Cn overestimated type I diabetes, 37% of type II diabetes being misclassified because insulin-treated. Classification Ccp observed 74 positive C-peptide patients, classified as type II, among whom 45 were insulin-treated. Only 3 patients were discordant for classification Cd and Ccp. Predictive value of "negative C-peptide" and "positive C-peptide" were 100% and 96% respectively. Multiple regression analysis of the Ccp classification was performed with the clinical criteria and showed very significant correlation with: age at the time of diagnosis of diabetes (AGE), maximal body mass index ever reached (BMI MAX) and delay between diagnosis and consistent insulin use (DI).(ABSTRACT TRUNCATED AT 250 WORDS)