Nonelective hospitalizations for urgent or emergent reasons are frequent events for patients with diabetes mellitus, and their occurrence is difficult to predict. A model for predicting nonelective hospitalizations is described. It is based on risk factors: prior visits to the emergency room, hypoalbuminemia, cardiomegaly, anemia, systolic hypotension, and hyperglycemia. To test the model, the authors conducted a prospective cohort study in which 429 ambulatory patients with diabetes mellitus were stratified into three risk levels for hospitalization and followed for two years. Patients in higher risk groups were more likely to be hospitalized (high risk, 58.1%; medium-risk, 40.2%; low risk, 26.6%, p less than 0.01) and had more hospitalizations per patient (1.47 vs. 0.80 vs. 0.46, p less than 0.01) and more hospital days per patient (14.6 vs. 8.6 vs. 5.3, p less than 0.01). When the two-year study period was divided into four six-month intervals, there was no significant difference across the four periods. This study demonstrates the validity of the model for predicting nonelective hospitalizations of patients with diabetes mellitus over time.