BACKGROUND: The delay in diagnosis and treatment of tuberculous meningitis (TBM) is a major factor in the high mortality observed with this pathology. The distinction between bacterial meningitis (BM) and TBM by clinical features alone is often impossible, and the available biological resources remain inadequate or inaccessible, especially in developing countries. We attempted to develop a simple diagnostic algorithm on the basis of clinical and laboratory findings that could be used as an early predictor of TBM in adult patients in Morocco. METHODS: We compared the clinical and laboratory features on admission of 508 adults in a medical intensive care unit in Morocco who satisfied diagnostic criteria for tuberculous (n=274) or bacterial (n=234) meningitis. Features independently predictive of TBM were modeled by multivariate logistic regression to create a diagnostic rule, and by a classification and regression tree (CART). RESULTS: Six features were predictive of a diagnosis of TBM: female gender, duration of symptoms, the presence of localizing signs, white blood cell (WBC) count, the level of serum sodium, and the total cerebrospinal fluid WBC count. The sensitivity for CART was 87% and for a score >7 was 88%; specificity was 96% and 95%, respectively. The internal validation was excellent for both diagnostic methods, with a receiver operating characteristic (ROC) area of 0.906 bootstrap samples for a score >7 and 0.910 for CART. CONCLUSIONS: The clinical and laboratory parameters identified in this study may help the clinician with the empiric diagnosis of TBM and could be used in settings with limited microbiological diagnostic support.