The rationale for use of benzodiazepine receptor antagonists is based on the so-called benzodiazepine pathogenetic hypothesis of hepatic encephalopathy (HE). To assess the efficacy of flumazenil, a specific benzodiazepine receptor antagonist, in a large and selected population of cirrhotic patients with severe HE, we conducted a double-blind, placebo-controlled, cross-over trial on 527 cirrhotic patients with HE grade III and IVa admitted to Intensive Care Units over a 5-year period; among them, 265 (132 of grade III and 133 of grade IVa) received flumazenil, whereas 262 (130 of grade III and 132 of grade IVa) received placebo. Treatment was begun within 15 minutes of randomization; the response to treatment was assessed by neurological score and by continuous electroencephalographic (EEG) recordings. Improvement of the neurological score was documented in 17.5% of grade III patients treated with flumazenil and in 14.7% of grade IVa patients, compared, respectively, with 3.8% and 2.7% of the patients of both groups treated with placebo. Improvements in EEG tracings were observed in 27.8% of grade III patients and in 21.5% of grade IVa patients, compared, respectively, with 5% and 3.3% of the patients of both groups treated with placebo. Benzodiazepines were detected in the serum of 10 patients (4 in grade III group and 6 in grade IVa group). Flumazenil is beneficial only in a selected subset of cirrhotic patients with severe HE; the applicability of this treatment to unselected patients with severe HE still remains to be determined.