In experimental animal studies the benzodiazepine antagonist (BZA) flumazenil reversed galactosamine-induced fulminant liver failure. Based on this observation, we treated three patients (two male, one female), mean age 41 yrs (34-52 yrs) with five episodes (ep) of hepatic encephalopathy (HE) with i.v. BZA over 24 h. The efficacy of BZA in reversing HE was evaluated. On admission all patients had elevated plasma ammonia levels. The patients were in coma stage II (n = 3 ep) and stage III (n = 2 ep); the modified Glasgow score was less than 10 (n = 1 ep), 10-15 (n = 2 ep), 15-20 (n = 2 ep). No evidence for benzodiazepine, barbiturate or opiate intake was found in the urine. Two patients with HE episode stage II showed an immediate recovery within 1 h, which continued without any therapy other than BZA. After 24 h BZA was given orally, 30 mg q.i.d. One patient with an HE episode stage II showed no effect over 12 h. After this time he progressed to HE stage III; BA therapy was stopped. After discontinuation of BZA the Glasgow score rapidly deteriorated. Twenty-eight hours later BZA was given again with i.v. mannitol (20%) and oral lactulose and Fordtran solution. The patient then recovered within the next day. Another patient with a HE episode stage III showed a short lasting improvement after BA bolus injection. Then impairment of the Glasgow score was noted again. Lactulose and Fordtran solution was added orally and mannitol (20%) was given i.v. The patient recovered within 24 h to HE stage II and after 72 h to normal. BZA seems to improve hepatic encephalopathy, especially at an early stage (II). It appears less effective in stage III, most of all when there is evidence of brain oedema. No specific side-effects were observed during BZA therapy.