This double-blind randomised study compared the analgesic efficacy, respiratory effects, side effects, and pharmacokinetic disposition of 24 hr lumbar epidural and intravenous infusions of the same dosage regimen of fentanyl (1.5 micrograms.kg-1 bolus then 1 microgram.kg-1.hr-1 infusion) in 50 patients after thoracotomy. Patients received either epidural fentanyl and intravenous normal saline, or epidural normal saline and intravenous fentanyl, for postoperative analgesia, after a standard low-dose alfentanil and isoflurane general anaesthetic. Visual analogue pain scores were lower in the epidural group (P < 0.05) only at two hours postoperatively, and there was no difference in the amount of supplementary morphine self-administered by patient-controlled analgesic pump. A mainly spinal analgesic effect probably occurred in the first few hours since fentanyl was not detectable in the plasma of patients in the epidural group until two hours after bolus injection; its concentration was less at that time than after intravenous injection (P < 0.05). Thereafter there was no difference in the plasma concentration profiles between the two groups. Seven patients in the epidural group and ten patients in the intravenous group received naloxone for PaCO2 > 50 mmHg, and one patient in the intravenous group had the infusions stopped because of PaCO2 elevation and somnolence. In patients who did not receive naloxone, the epidural route produced better analgesia throughout the study period (P < 0.01).(ABSTRACT TRUNCATED AT 250 WORDS)