The aim of this study was to validate a diagnosis model that provides pABM, the probability of bacterial versus viral meningitis, based on four parameters collected at the time of first lumbar tap: cerebrospinal fluid protein level, cerebrospinal fluid polymorphonuclear cell count, blood glucose level, and leucocyte count. The model was evaluated prospectively as an aid to therapeutic decision-making in 109 consecutive patients with acute meningitis and negative cerebrospinal fluid Gram stain. In each case pABM was computed before a therapeutic decision and three diagnoses were established successively: (i) clinical evaluation, i.e. before pABM computation (bacterial meningitis, viral meningitis, or meningitis of undetermined origin); (ii) computation of pABM (viral meningitis if pABM< 0.1, bacterial meningitis otherwise); and (iii) determination of definitive diagnosis (bacterial meningitis: positive cerebrospinal fluid culture; viral meningitis: negative cerebrospinal fluid culture, no other aetiology and no treatment; meningitis of undetermined origin: cases fitting neither of the first two diagnoses). The computed diagnosis was viral meningitis in 78 of the 80 cases diagnosed definitively as viral meningitis, and bacterial meningitis in four of the five cases diagnosed definitively as bacterial meningitis. Negative and positive predictive values and accuracy of the model were 98.7%, 66.7%, and 96.5%, respectively. The clinical diagnosis was undetermined in 22 cases, 15 of which were diagnosed definitively as viral cases; in all of these 15 cases, the computed diagnosis was viral meningitis, leading the physician to refrain from starting antibiotics in all of them. The results confirm that the model evaluated is reliable and aids in the identification of patients in whom antibiotics can be safely avoided.