Ability to prevent the development of diabetic complications is related to the metabolic control of the disease. A risk predictive model would allow to identify, at the time of diagnosis, patients at risk of inadequate control.Objective: to determine the risk predictive variables that are present at the begining of the illness, to develop a model to predict morbidity in IDDM children. Materials and methods: IDDM pediatric subjects controlled during at least 3 years in our Hospital's Nutrition Service were studied. No subject had associated pathologies. Their families were surveyed to obtain information about socioeconomic status, family structure, health care, parental education and biological aspects, all variables referred to the begining of the illness. Patients were classified into two groups:“good” (GC) and “bad” controls (BC) according to three parameters: average value of HbAlC, delta Height in z score and number of admissions due to ketoacidosis. To homogenize the data these parameters were obtained for the first three years of disease. Analysis was performed with the SPSS program, using univariate and multiple logistic regression. Student's two tailed t test for independent samples was used for continuous variables and X2 test for categorical groups. Multiple stepwise logistic regression analysisi was used to investigate the relative contribution of independent predictor factors. Results: 88 patients were studied. According to a pre-established score the percentage of patients with GC was 48% vs 52% with BC. There was no statistically significant difference for the age of onset between groups. Of the 26 variables surveyed, the univariate analysis showed significant association with metabolic control: 1) to have an attending pediatrician (p<0.01): 2) to have received adequate medical care during the first year of life (p<0.01), 3) father's employment (p<0.03): 4) father(p<0.02) and mother instrucction levels (p<0.04): 4) parents living together (p<0.03) and 5) presence of both parents during diabetologic education (p<0.03). In multiple logistic regression analysis the following associations were significant 1) to have an attending pediatrician: 2) have received adequate medical care during first year of life and 3) parents living together. Conclusions: identification at the time of diagnosis of variables significantly associated with “bad” control will allow us to apply a predictive model of morbidity. These model may be a contributory strategy to identify the patients with high risk to develop complications.