This randomized, double-blind study of 40 patients was designed to determine if the predominant analgesic effect of extradural fentanyl is mediated by a direct spinal action or an indirect systemic one. After Caesarean section, postoperative analgesia was provided for 24 h by patient-controlled extradural analgesia (PCEA group) or by patient-controlled i.v. analgesia (PCIVA group). Both groups received a bolus dose of fentanyl 20 micrograms with a 10-min lockout interval. In the PCIVA group, nine patients stopped early (compared with none in the PCEA group) because of inadequate analgesia. Mean visual analogue pain scores (0-100 mm) at 8 and 12 h were lower for PCEA (23 (sd 13) mm at rest, 31 (23) mm on coughing) than for PCIVA (50 (25) mm at rest, 67 (24) mm on coughing) (P < 0.0005). The mean dose of fentanyl self-administered between 4 and 8 h was lower in the PCEA group (38 (sd 30) micrograms h-1) compared with the PCIVA group (59 (27) micrograms h-1) (P < 0.05). Our results support the hypothesis that the predominant analgesic effect of extradural administration of fentanyl is mediated by a direct spinal action rather than an indirect action from systemic absorption.