1. One hundred and two consecutive patients undergoing upper gastrointestinal endoscopy were randomised to be sedated with either intravenous diazepam (Diazemuls-Kabi Vitrum) or intravenous midazolam (Hypnovel-Roche). It was assumed that midazolam was likely to be approximately twice as potent as diazepam on the basis of previous work. 2. All patients had an ear oximeter attached throughout the procedure to record continuously their level of oxygen saturation. 3. All 102 patients had pre-endoscopy respiratory function tests measured and 100 wore an induction plethysmograph vest to allow continuous estimation of respiratory rate and excursion. The plethysmograph was calibrated using a pneumotachygraph, so baseline, post-injection and post-endoscopy minute volumes could be estimated. 4. The age, sex ratio and pre-endoscopy respiratory function tests of the 51 patients given intravenous diazepam in a mean dose (s.d.) of 11.5 (5.8) mg over a mean of 3.4 (0.9) min) were similar to that of the 51 patients sedated with intravenous midazolam (mean dose 6.0 (2.8) mg over 3.3 (0.9) min. 5. Both drugs significantly reduced minute volume (P less than 0.001) and oxygen saturation (P less than 0.001). Midazolam appeared to produce slightly greater hypoxaemia with 57% having falls in oxygen saturation of greater than 2.5% compared with only 35% given an equivalent dose of diazepam. 6. Ventilation was still less than baseline when re-checked some minutes after removal of the gastroscope. The speed of recovery appeared faster after diazepam sedation which is in contrast to its longer pharmacological half-life.(ABSTRACT TRUNCATED AT 250 WORDS)