Twenty-nine patients, ranked ASA 1 or 2, scheduled for diagnostic gastroduodenal fibroscopy were randomly allocated into three treatment groups: intranasal midazolam (IN) 0.15 mg.kg-1; intramuscular midazolam (IM) 0.15 mg.kg-1, and placebo (P). Sedation and the fibroscopy were all carried out by the same anaesthetist and fibroscopist. Efficiency and tolerance of the method were assessed by monitoring the heart rate, systolic and diastolic arterial blood pressure, arterial oxygen saturation (pulse oximetry), the degrees of sedation, anxiety, and anterograde as well as retrograde amnesia. The patient's and endoscopist's reactions were also considered. Patients in group P were older than those of the other two groups (p less than 0.01). Sedation was more important and patients less anxious in groups IN and IM than in group P. Three patients in group IM had retrograde amnesia. There were no significant differences between the degrees of anterograde amnesia in the three groups. At no time during the study were there any significant differences in SpO2 between and within groups. Four out of the 9 patients in group P had a bad opinion of their experience, as compared with two out of the 20 in the midazolam groups; the difference was not statistically significant. Three hours after the procedure, all patients were fully awake. Intranasal administration of midazolam therefore seemed to be an interesting alternative for gastroduodenal fibroscopy, because it was simple, non traumatic, well tolerated, and did not result in arterial desaturation.