The effectiveness of the benzodiazepine antagonist, flumazenil, was evaluated in a randomized double-blind clinical study in which diazepam 0.2 mg kg-1 or midazolam 0.1 mg kg-1 was used for i.v. sedation. We studied 120 day-case patients undergoing gastroscopy and treated with either flumazenil 0.1-2 mg or placebo after the procedure. Psychometric assessment of four aspects of recovery over a 3-h period showed that flumazenil attenuated the sedative effects of the benzodiazepines, but did not antagonize the sedation completely. For patients sedated with diazepam, there was a significant effect of flumazenil on speed of motor co-ordination after 90 min (P less than 0.01), and for those given midazolam a similar effect was found at 20 min (P less than 0.01). However, after 3 h all four groups of patients had not returned to baseline performance in accuracy of motor co-ordination (P less than 0.01) and cortical arousal (P less than 0.05), and the two groups sedated with diazepam still displayed memory deficits (P less than 0.05). Flumazenil did not attenuate the subjective experience of sedation as measured by visual analogue scales. These results indicate that sedation is multidimensional, differentially affecting the hierarchy of cognitive functions. In day-cases, antagonism of benzodiazepine sedation with flumazenil would not hasten the safe discharge of patients.