Ten long-term users of benzodiazepines (average daily dose, 20 mg of diazepam or equivalent) who had experienced problems in withdrawing from the drugs were given an i.v. challenge with either the benzodiazepine antagonist flumazenil (1 mg injected over 30 s) or placebo (vehicle solution) in a randomized double-blind design. There were no 'pseudo withdrawal' responses to either single-blind or double-blind placebo injections, whereas flumazenil produced dramatic panic reactions in all four subjects tested, followed by characteristic benzodiazepine withdrawal symptoms. There were also small but significant rises in pulse rate and blood pressure, but no change in serum cortisol. Flumazenil-induced panic could not be entirely accounted for by a past or present diagnosis of panic disorder, and did not seem to be related to previous withdrawal problems, present benzodiazepine dosage, or to the severity of withdrawal symptoms precipitated by flumazenil in the same challenge test. Attempts to reduce benzodiazepine intake over the next 3 weeks tended to be more successful in the flumazenil group. The results are discussed with reference to possible changes in the GABA-benzodiazepine system in long-term benzodiazepine users.