A large number of biomarkers have been discovered over the past few years using proteomics techniques. Unfortunately, most of them are neither specific nor sensitive enough to be translated into in vitro diagnostics and routine clinical practice. From this observation, the idea of combining several markers into panels to improve clinical performances has emerged. In this article, we present a discussion of the bioinformatics aspects of biomarker panels and the concomitant challenges, including high dimensionality and low patient number and reproducibility.