Various techniques and drug regimes for thoracic paravertebral block (PVB) have been evaluated for post-thoracotomy analgesia, but there is no consensus on which technique or drug regime is best. We have systematically reviewed the efficacy and safety of different techniques for PVB. Our primary aim was to determine whether local anaesthetic (LA) dose influences the quality of analgesia from PVB. Secondary aims were to determine whether choice of LA agent, continuous infusion, adjuvants, pre-emptive PVB, or addition of patient-controlled opioids improve analgesia. Indirect comparisons between treatment arms of different trials were made using metaregression. Twenty-five trials suitable for metaregression were identified, with a total of 763 patients. The use of higher doses of bupivacaine (890–990 mg per 24 h compared with 325–472.5 mg per 24 h) was found to predict lower pain scores at all time points up to 48 h after operation (P=0.006 at 8 h, P=0.001 at 24 h, and P<0.001 at 48 h). The effect-size estimates amount to around a 50% decrease in postoperative pain scores. Higher dose bupivacaine PVB was also predictive of faster recovery of pulmonary function by 72 h (effect-size estimate 20.1% more improvement in FEV1, 95% CI 2.08%–38.07%, P=0.029). Continuous infusions of LA predicted lower pain scores compared with intermittent boluses (P=0.04 at 8 h, P=0.003 at 24 h, and P<0.001 at 48 h). The use of adjuvant clonidine or fentanyl, pre-emptive PVB, and the addition of patient-controlled opioids to PVB did not improve analgesia. Further well-designed trials of different PVB dosage and drug regimes are needed.