A prospective, randomly controlled study was conducted to test the effect of continuous flumazenil infusion in preventing complications arising from severe benzodiazepine (BZ) intoxication. Patients who were believed to be suffering benzodiazepine intoxication and whose Glasgow Coma Scale (GCS) score was below 10 were enrolled after showing a clear-cut response to flumazenil 0.5 or 1 mg (an improvement by 4 or more on the GCS). The patients were consecutively enrolled and randomized into two groups: a continuous infusion group (CI, n = 50) who were immediately given flumazenil 0.5 mg/h for 5 hours, and a control group (CIN, n = 50). Age, sex, incidence of underlying disease, GCS score at several time points, and complication rate were compared in the two groups. Although the CI group had a higher GCS score at most time points, the complication rate did not significantly differ between the two groups (14 of 36 in the Cl group v 12 of 38 in the CIN group, P = .684). A greater incidence of underlying disease and an older age seemed to contribute to the higher complication rates in both groups. Several patients (in both groups) resedated into deeper coma after showing an initial response to flumazenil or after the cessation of flumazenil infusion. For severe BZ intoxication, treatment with flumazenil infusion should still be considered skeptically and should not be recommended as routine management. BZ-intoxicated patients with an underlying disease, an older age, and resedation into a deep comatose state after showing an initial response to flumazenil should be treated in an intensive care unit.