Meta-analysis has been defined as a study and "statistical analysis which combines or integrates the results of several independent studies." Included in this definition are other terms, such as systematic overviews, pooling data, pooling study results, and quantitative literature reviews. Like any study, the questions being asked will influence the design and the method of analysis of the meta-analysis. Since a meta-analysis is a study based on a literature review, it is inherently observational rather than experimental in nature. This idea is supported by the fact that the meta-analyst has limited control over the availability of studies or the information collected and reported in the individual studies. Meta-analysis has been applied to clinical trials and epidemiology. At first glance the potential for bias appears greater in epidemiology than in clinical trials. But this may depend on the question being asked. If randomized clinical trials are limited to improving an estimate of effect or testing a hypothesis in a relatively homogeneous set of effect sizes, the clinical trial will tend to be less prone to bias than a comparable set of epidemiologic studies. In this context, the issue of combinability may dominate the meta-analysis. We refer to this type of meta-analysis as an "analytic" meta-analysis. On the other hand when the goal is to resolve controversy, or pose and answer new questions the main concern of the meta-analysis is to explain the variation in the effect sizes. We refer to this application of a meta-analysis as an "exploratory" meta-analysis. In this second type of meta-analysis the characteristics of the different studies become the focus of the analysis. This leads to the idea that protocols for a meta-analysis should reflect its goals and how the results are to be used. Finally, we will consider whether there is a role of meta-analysis in the field of drug development.