Forty-five patients were allocated randomly to receive either a single intrathoracic block of four intercostal nerves, a continuous thoracic extradural infusion or a continuous paravertebral infusion of bupivacaine. Patients were allowed additional i.v. boluses of morphine via a PCA device. Segmental spread of pinprick analgesia was comparable in the groups for up to 20 h. Up to 2 h after the block, plasma concentrations of bupivacaine were greater in the intercostal group and there was large interindividual variation. There were no significant differences between the groups in pain, morphine consumption, respiratory function or adverse events. Moderate to severe respiratory depression was detected in 14 patients more than 2 h after operation.