We analyzed data from the records of 422 patients with acute bacterial or viral meningitis. A cerebrospinal fluid (CSF) glucose level less than 1.9 mmol/L, a CSF-blood glucose ratio less than 0.23, a CSF protein level greater than 2.2 g/L, more than 2000 × 106/L CSF leukocytes, or more than 1180 × 106/L CSF polymorphonuclear leukocytes were individual predictors of bacterial infection with 99% certainty or better. Although any one of these tests could rule in bacterial meningitis with high probability, none could rule it out. To better predict whether a patient has bacterial vs viral infection, we developed a logistic multiple regression model using CSF-blood glucose ratio, total polymorphonuclear leukocyte count in CSF, age, and month of onset. This proved highly reliable when validated in an independent test sample, with an area under receiver operating characteristic curve of 0.97. The model should allow physicians to differentiate between acute viral and acute bacterial meningitis with greater accuracy.