Using a double-blind procedure, 68 patients with putative benzodiazepine dependence were randomly allocated to one of three groups given lorazepam (n = 22), diazepam (n = 23) or bromazepam (n = 23) in doses equivalent to those of the patients' original benzodiazepine. After four weeks the dosage was reduced in 25% quantities until no further benzodiazepines were taken. A total of 23 patients dropped out during the study, ten on lorazepam (one of whom committed suicide), seven on diazepam and six on bromazepam. There were few differences in withdrawal symptoms between the three groups but, despite the higher dropout rate, these symptoms were somewhat less marked in the lorazepam group. Withdrawal symptoms were greater in patients who had taken a benzodiazepine for greater than 5 years and were most marked in those with personality disorders, predominantly dependent ones.