Fifty-three critically ill infants and children received midazolam as sedation in a regional intensive care unit. Assessment of the level of sedation was carried out at regular intervals on withdrawal of midazolam. Forty-nine patients were fully alert within 4 h of midazolam being stopped. Four patients took from 6 h to 1 week to become fully alert. Four patients had abnormal behaviour highly suggestive of midazolam withdrawal. The onset of abnormal behaviour was within 12 h of discontinuation of midazolam. The duration of the abnormal behaviour ranged from 3 h to 1 week. One child had a paradoxical reaction to midazolam. The overall incidence of adverse effects to midazolam in the patients studied was 17%. No adverse effects were observed in infants; all adverse effects were observed in children. We have shown that it is possible to prospectively study the toxicity of sedatives in critically ill infants and children.