Objective To assess the efficacy of high-dose chemotherapy versus moderate-dose chemotherapy in the treatment of osteosarcoma. Methods We searched MEDLINE, EMbase, OVID database, CBMdisc, Cochrane CENTRAL Register of Controlled Trials in The Cochrane Library, and handsearched Journal of Chinese Oncology, Journal of Chinese Clinical Oncology and Tumor. The search time was updated to Feburary 2006.The quality of the included studies was evaluated by two reviewers and meta-analyses were performed on the results of homogenous studies. Results Four studies involving 937 participants with primary, high-grade and non-metastatic extremity osteosarcoma were included. All the included studies were judged to be inadequate at reporting randomization and blinding, only one reported allocation concealment. All included studies reported the number of withdrawals and the reasons for these. The meta-analyses showed that there were no significant differences in 5-year event free survival (EFS) (RR 1.10, 95% CI 0.96 to1.25), 5-year overall survival (OS) (RR 1.08, 95% CI 0.97 to1.20), local recurrence rate (RR 0.92, 95% CI 0.54 to 1.57), proportion of good histological response (RR 0.93, 95% CI 0.81 to 1.07), proportion of limb salvage [RR 0.97, 95% CI 0.92 to 1.02) between the high-dose group and the moderate-dose group. The 5-year EFS of the good histological response group was significantly higher than in the poor histological response group [OR 2.45, 95% CI 1.76 to 3.39,Plt;0.00001 ). Conclusions No advantage is shown for high-dose chemotherapy over moderate-dose chemotherapy in 5-year EFS, 5-year OS, local recurrence rate, proportion of good histological response and proportion of limb salvage. Histological response to preoperative chemotherapy is an independent prognosis factor for osteosarcoma. Due to the potential risk of selection bias, performance bias and publication bias, the evidence is not b enough to judge whether high-dose chemotherapy is better than moderate-dose chemotherapy in the treatment of osteosarcoma. Our conclusion suggests that large-scale randomized trials should be performed.