Patients may find bronchoscopy without sedation unpleasant. There is some evidence that patient satisfaction correlates with amnesia for the procedure. For several years we have used doses of midazolam sufficient to put patients lightly asleep hoping to produce complete amnesia. We looked at practical aspects of this technique over a 2-year period. We studied 337 consecutive patients. They were 219 men and 118 women of mean age 63 +/- 12.4 (SD). Sixty-seven patients were aged 75 years or over and the eldest was 86. Sixty-three patients were already hospital inpatients but the remainder were seen as day cases. Midazolam was given by slow i.v. injection over several minutes until the patient was judged to be lightly asleep. Patients were given supplemental oxygen (3 l min-1) and monitored by ECG and pulse oximetry. A note was made of the time at which they awakened, defined as when nursing staff felt the patients were awake enough to have a cup of tea and toast. Patients were asked if they had any memory of the procedure both on awakening and when seen a few days later to discuss the results. The procedures were carried out in a well-staffed Day Case Unit with a recovery area. The mean dose of midazolam used was 10.8 mg (mean +/- SD = 0.16 +/- 0.095 mg kg-1). The midazolam was given over a median of 4 min (range 1-15 min). Patients took 59 +/- 45 min (mean +/- SD) to wake up. Twenty-eight patients were given flumazanil to reverse the sedation (11 for concern over bleeding following biopsies, three for desaturation during and three after procedure, four as they were frail, two as they were restless, two as they were hypotensive after procedure and three for miscellaneous reasons). Only nine patients could remember any part of the procedure. Incremental doses of midazolam given slowly until patients are lightly asleep almost invariably produce complete amnesia for bronchoscopy. This is a safe technique but patients need careful monitoring and may require reversal of sedation with flumazanil.