The separation of proteins with the aim of discovering surrogate biomarkers defining differences between various stages of biological materials is the core occupation of every project in Proteomics. There are numerous recent publications suggesting a wide array of separation technologies, ranging from 2-DE, MS-linked LC, CE or chip-based surface-enhanced laser desorption ionization claiming to be useful for this purpose, and addressing the urgent clinical, diagnostic or toxicological needs for such surrogates. However, many potential biomarkers emerging from proteomic studies did not survive validation in, for example, large-scale clinical studies or simply independent experiments, and at the same time being tested in settings with case numbers bigger than perhaps a few hundreds. The major problems of protein biomarkers are associated with the huge dynamic range of possible concentrations and the ever-increasing number of molecular species due to post-translational modifications. In particular, the chemical diversity of the latter imposes a necessity of improved resolution of separation technologies, because otherwise the crucial quantitative information is lost in pools of poorly resolved peptides. Here, we present and analyze some examples of successful developments of protein biomarkers, and show the prerequisites and necessary considerations while moving protein candidates from purely descriptive phenomena to a stage of validated surrogate biomarkers. This includes a detailed discussion of requirements regarding resolution of initial separation techniques, linear dynamic range and statistics of differential quantification, but also the subsequent clinical validation, testing the biomarker in clinical settings and using large numbers of patient samples.