We propose a novel R package, called COSMONET (COx Survival Methods based On NETworks), which implements (a) three different screening techniques to reduce the initial dimension of the data from a high-dimensional space to a moderate scale, (b) a network-regularized Cox-regression algorithm to select the gene signatures, (c) various approaches to calculate an optimal cut-off on the prognostic index (PI) in order to divide the patients in high- and low-risk group and (d) a prediction analysis for patients’ risk based on the evaluation of PIs.