The usefulness of intravenous patient-controlled analgesia (PCA) with opioids for postoperative analgesia is not well defined. We systematically searched (MEDLINE, EMBASE, Cochrane Library, bibliographies, any language, to January 2000) for randomised trials comparing opioid-based PCA with the same opioid given intramuscularly, intravenously, or subcutaneously. Weighted mean differences (WMD) for continuous data, relative risks (RR) and numbers-needed-to-treat (NNT) for dichotomous data were calculated with 95% confidence intervals (CI) using fixed and random effects models. Data from 32 trials were analysed: 22 (1139 patients) were with morphine, five (682) with pethidine, three (184) with piritramide, one (47) with nalbuphine and one (20) with tramadol. In three morphine and one pethidine trial (352 patients), more patients preferred PCA (89.7% vs. 65.8%, RR 1.41 (95%CI 1.11 to 1.80), NNT 4.2). Combined dichotomous data on pain intensity and relief, and the need for rescue analgesics from eight morphine, one pethidine, one piritramide, and one nalbuphine trial (691 patients), were in favour of PCA (RR 1.22 (1.00 to 1.50), NNT 8). In two morphine trials (152), pulmonary complications were more frequently prevented with PCA (100% vs. 93.3%, RR 1.07 (1.01 to 1.14), NNT 15). There was equivalence for cumulative opioid consumption, pain scores, duration of hospital stay, and opioid-related adverse effects. These trials provide some evidence that in the postoperative pain setting, PCA with opioids, compared with conventional opioid treatment, improve analgesia and decrease the risk of pulmonary complications, and that patients prefer them.