The authors discuss existing methods of meta-analysis, and show that meta-analysis based on individual patient data (IPD meta-analyses) are the most reliable. Taking the example of four successive meta-analyses of clinical trials in advanced colorectal cancer, they illustrate the possibilities of IPD meta-analysis. They conclude that meta-analyses are a powerful tool, not only to confirm small differences between treatment modalities, but also to generate hypothesis, to perform exploratory subgroup analysis, and to study potential surrogate endpoints.