From the medical records of 238 intensive care unit (ICU) patients who had infections with gram-negative pathogens commonly associated with serious illness, we developed a predictive score of clinical risk factors for seizures. To evaluate the predictive ability of this score, we applied it to a separate population of 645 seriously ill hospitalized patients with similar gram-negative infections who were in antibiotic clinical trials. The patients at highest risk were classified into one of the following three categories: (a) patients with major central nervous system (CNS) insults (CNS surgery, hemorrhage, infection, or other lesion within 1 month before hospital admission or any history of CNS neoplasia), (b) patients with a predisposing factor (renal impairment or a history of seizures) plus a precipitating factor (anoxic encephalopathy/coma or an acute hypotensive episode), and (c) patients with both renal impairment and a history of seizures. Receiver operating characteristic (ROC) curves were calculated in each of the two populations. The area under the ROC curve (AUC) represents the probability that the score would rank a randomly chosen patient who subsequently had a seizure as having had a greater prior level of seizure risk than a randomly chosen patient who did not experience a seizure. The AUC was 0.87 (SE = 0.05) for the original population used to develop the score and 0.81 (SE = 0.04) for the population used for the validation study. The clinical risk score, based on readily available information, provides a useful means to identify among seriously ill infectious disease service patients, those who are at highest risk for seizures. It also serves as a baseline for evaluating the non-drug-related risk factors for seizures in patients treated with antibiotics.