Objective-To assess the diagnostic value and safety of the benzodiazepine antagonist flumazenil in patients with coma ofunclear origin with suspected poisoning.Design-Double blind, placebo controlled, randomised study.Setting-Intensive care unit at a major teaching hospital.Patients-105 Unconscious adults admitted consecutively with suspected drug overdosage during 18 months from a total of 362 cases of poisoning. Exclusion criteria were pregnancy, epilepsy, obvious poisoning with drugs identified unequivocally from information from relatives or others as other than benzodiazepines, and coma score >10 on a scale graded from 4 to 20. Patients were allocated randomly to receive flumazenil (21 men and 32 women) or placebo (25 men and 27 women).Interventions -Intravenous injection offlumazenil (10 ml, 0-1 mg/ml) or placebo (10 ml vehicle alone) given double blind over three minutes.Main outcome measures -Serum and urine concentrations ofbenzodiazepines, antidepressants, and several other agents; blood gas tensions; standardised evaluation on admission and five minutes after the injection by means of coma scale score and urgent diagnostic or therapeutic interventions indicated according to the history and clinical examination; standardised interview after the injection to try to ascertain further information; and adverse reactions.Results-Benzodiazepines were found in the serum in 36 of the 53 patients in the flumazenil group and in 37 of the 52 who received placebo. The average coma scale score increased significantly after injection in the flumazenil group (6-4 v 12-1, p<0-001) but not in the placebo group. In the flumazenil group several interventions were rendered unnecessary by the injection: gastric lavage and urinary catheterisation (19 patients each), intubation (21), artificial ventilation and computed tomography of the brain (three patients each), blood culture and lumbar puncture (one patient each), and electroencephalography (two). In the placebo group the indications for these procedures did not change in any patient after injection. The 95% confidence interval for the difference in reduction of the frequency ofindications for gastric lavage after injection between the two groups was 21% to 51%, that for intubation 25% to 55%, and that for urinary catheterisation 21% to 51%. In the flumazenil group 21 patients gave valuable information on their drug ingestion within 10 minutes after injection compared with only one in the placebo group (p<0-001). Nine adverse reactions were recorded in the flumazenil group, eight of which were graded as mild and one severe.