Midazolam in doses of 0.21-0.46 mg/kg was given to four insomniacs for 27 days, and to a fifth insomniac for an additional 124 days, to assess its short, intermediate, long-term and long-standing efficacy. Automatic analysis of the polygraphic sleep recordings was carried out employing real-time signal processing (based on linear prediction), and an off-line sequential fuzzy clustering. Total sleep length did not vary significantly during medication, whereas other efficacy parameters such as sleep onset latency, wake time after sleep onset and number of nightly awakenings decreased significantly throughout the period of active medication. In one patient, midazolam was still effective after 151 days of administration. The ultrashort action (probably 3-4 h) of midazolam could be demonstrated by the distribution of the efficacy parameters, dividing the night into 2-h periods, by the unchanged total sleep length during medication compared with baseline and by the lack of any subjective clinical symptoms of hangover. The quantitative analysis of the signals showed that, although a decrease in the total amount of slow-wave sleep during long-term medication and withdrawal was detected, the total relative power in the delta band increased during drug administration. This was interpreted as redistribution of the delta activity during active medication. A lengthening of REM sleep was found during active medication probably attributable to intranight rebound. This was not followed by any clinical symptom of clinical REM rebound.(ABSTRACT TRUNCATED AT 250 WORDS)