Abstract: Objective To introduce the principle of the multivariate meta-analysis and its’ implementation by the mvmeta package in R language. Methods The randomized trials of non-surgical treatment of first bleeding in cirrhosis using beta-blockers, sclerotherapy and control group was used as an example. The log-scale of first bleeding risk of OR values that two therapies compared control group were calculated separately and regarded as multivariate, and then multivariate meta-analysis was used by mvmeta package in R language. Results Multivariate meta-analysis showed: OR and their 95% confidence interval of beta-blockers, sclerotherapy compared the control group were 0.508 9 (0.270 5, 0.957 1) and 0.552 2(0.318 4, 0.957 7) respectively. The first bleeding risk of the two therapies are less than 1. The difference is statistically significant. The correlation of between-study is 0.436 5. Conclusion When meta-analysis has the multiple endpoints or groups, multivariate meta-analysis takes multiple endpoints and the between-study correlation into account simultaneously, making the progress more reasonable and the result of multivariate meta-analysis more accurate and credible.