Many genome-scale studies in molecular biology deliver results in the form of a ranked list of gene names, accordingly to some scoring method. There is always the question how many top-ranked genes to consider for further analysis, for example, in order creating a diagnostic or predictive gene signature for a disease. This question is usually approached from a statistical point of view, without considering any biological properties of top-ranked genes or how they are related to each other functionally. Here we suggest a new method for selecting a number of genes in a ranked gene list such that this set forms the Optimally Functionally Enriched Network (OFTEN), formed by known physical interactions between genes or their products. The method allows associating a network with the gene list, providing easier interpretation of the results and classifying the genes or proteins accordingly to their position in the resulting network. We demonstrate the method on four breast cancer datasets and show that 1) the resulting gene signatures are more reproducible from one dataset to another compared to standard statistical procedures and 2) the overlap of these signatures has significant prognostic potential. The method is implemented in BiNoM Cytoscape plugin (http://binom.curie.fr).