Objective To investigate the overall probability of error in preparing and administering intravenous medicines; to identify at which stage of the process an error is most likely to occur; and to determine the impact of error correction on the error probability. Design Systematic review and random-effects Bayesian conditional independence modelling. Methods Medline and EMBASE were searched for studies on intravenous medicines. The error rates of each stage were extracted. These, expert estimates, and error rates from generic tasks, were used in a Bayesian conditional independence model to find error 'hot-spots.' The main outcome measure was the probability of at least one error occurring during intravenous therapy. Results Nine published studies were identified for inclusion in the systematic review and meta-analysis. The overall probability of making at least one error in intravenous therapy was 0.73 (95% credible interval (CrI) 0.54 to 0.90). If error-checking was introduced at each stage of the process, the overall rate fell to 0.22 (95% CrI 0.14 to 0.31). Errors were most likely in the reconstitution step. Removing the reconstitution step by providing preprepared injections would reduce the overall error rate to 0.17 (95% CrI 0.09 to 0.27). Conclusions Intravenous therapy is complex and error-prone. Error-checking at each stage could reduce the error probability. The use of preprepared injections may help by eliminating errors in the reconstitution of drug and diluent. However, it will be important to ensure that benefits are not outweighed by practical disadvantages such as an increase in selection errors.