OBJECTIVES: In this presentation we summarise statistical methods for meta-analysis when a direct comparison between treatment effects is impossible, inadequate, or inappropriate. METHODS: Detailed descriptions are presented, and these are appraised per se and in relation with conventional meta-analysis methods. The main methods can be summarised as follows: weighted mean difference of relative effect measures (e.g. mean difference, log-odds-ratio, log-relative-risk and log-hazard-ratio) and meta-regression of relative effect measures, both of which are based on traditional meta-analysis approaches, and weighted Bayesian regression models, which are more flexible and are simple to implement in freely available software. RESULTS: Using health outcomes research examples for illustration in each case, we describe common methodology issues arising from use of these methods, such as when small numbers of trials are analysed, when unequal trial sizes are included and when excess variability between trials (or heterogeneity) is encountered. CONCLUSIONS: For the methods considered, we offer possible solutions, make recommendations for their use and point out situations in which caution should be exercised.