In the present work it is reviewed the the statistical tools available for the biomarker discovery process. Resampling techniques, feature selection, classifiers and metric to evaluate the performance of the biomarkers is reviewed. All the above concepts are tested, applying them into an appropriate work-flow, in a real prostate cancer study. The aim is to find biomarkes to be able to classify men depending the aggressiveness of the cancer after an intervention, from the analysis of the miRNA expression obtained from qPCR. Three different comparisons are assessed, and in each one a total of 20 models are compared to find the best. The R/Bioconductor CMA package is used to perform the analysis. Potential biomarkers are found, although further studies are recommended before its commercialization or other use.