A historical cohort was used to assess the ability of clinical features and laboratory values recorded at the time of initial diagnostic investigations to predict nondiagnostic hospital admissions in the first 3 months following the diagnosis of Crohn's disease. Data were abstracted from the medical records of 225 eligible patients at primary and secondary care level whose disease was diagnosed between 1977 and 1985. The total study group was randomly divided into two groups (group 1, n = 112; group 2, n = 113). Discriminant analysis was performed on data of patients in group 1. The resulting predictive model was then cross-validated on data of patients in group 2. The variables entered into the predictive model were identified using bivariate analysis. Results show that presence of abdominal mass, body temperature, absolute basophil and lymphocyte counts, aspartate aminotransferase and blood urea nitrogen serum levels, and place of residence (urban, rural, or out of province) were the most useful variables for predicting hospitalization in the first 3 months (P for model = 0.0010; accuracy = 88%). Cross-validation on group 2 showed an accuracy of 80%, a positive predictive value of 62%, and a negative predictive value of 84%. This predictive model could be useful for counseling purposes on the primary or secondary care levels.