A meta-analysis of trials addressing the same question is a quantitative synthesis of their results. Such a synthesis is useful when the overall result of these trials does not produce a clear answer. In oncology, the moderate effect of most available treatments explains the widespread use of meta-analyses. They are one of the way to reach a sufficient power. A meta-analysis provides a test for the efficacy of the treatment under study, and an estimation of this efficacy. To perform a meta-analysis, one must identify all the randomised trials addressing the question, whether their results have been published or not. The quality of a meta-analysis depends largely on the completeness of the collection of trials. A meta-analysis can be based on summary data or it can be based on individual patient data which allows intensive data checking, data corrections and update, and an intent to treat analysis. Individual patient data meta-analysis is the reference method. It is particularly useful for controversial questions, or when a small benefit is expected or when long-term follow-up is important, all these situations being frequent in oncology. It allows extensive analyses, for instance the study of the variation in treatment effect according to patient's and trials characteristics. Detailed verification of the data are essential since the inclusion of ill designed and ill conducted trials can only lead to a bad meta-analysis. Close collaboration with investigators is needed in this type of meta-analysis leading to large and rapid diffusion of the results. Meta-analyses and large-scale trials are complementary tools for treatment evaluation in oncology. Lastly, the large data bases of individual patient data meta-analyses allow subsidiary studies such as validation studies of surrogate endpoints or studies of the predictive value of tumour markers.