A double-blind randomized study was performed to evaluate the efficacy and safety of flumazenil, a benzodiazepine antagonist. The study comprised 52 patients admitted to an intensive care unit because of suspected pure or mixed benzodiazepine poisoning. The degree of consciousness was assessed according to a modified Glasgow Coma Scale (MGCS), graded from 4 to 20, immediately before and at consecutive intervals after an i.v. injection of flumazenil or placebo. If there were no clear signs of arousal within 5 min after the blind injection, an injection of flumazenil was given in open design. Five minutes after the blind administration of active drug the MGCS score was increased by an average of 7.4 (p less than 0.001) in the flumazenil group. In the placebo group the average MGCS score of 7.8 on admission was not significantly increased by placebo, but 5 min after flumazenil injection it had increased by 7.3 to 15.1 (p less than 0.001). Patients who had ingested both alcohol and benzodiazepines were aroused as promptly and clearly (MGCS +8.8; p less than 0.001) as those who had taken benzodiazepines alone (MGCS +8.4; p less than 0.001). The patients poisoned with a combination of benzodiazepines and other hypnotic drugs responded less, but still highly significantly (MGCS +5.8; p less than 0.001). Flumazenil was well tolerated and the safety of the antidote seems acceptable. It is concluded that flumazenil can facilitate differential diagnosis and that it is an effective tool in the treatment of drug overdosage when benzodiazepines are involved.