Project portfolio selection is very complex. Often the multiple criteria used to assess projects are hierarchically structured and show interaction. Most of these criteria are also used at portfolio level to assess the supported projects as a whole; however, additional criteria must be considered to define the conformity of the decision maker with the portfolios. Here, we present a novel method to comprehensively address all these characteristics of the problem. The method uses a generalization of the outranking approach and value functions to aggregate the criteria scores, and evolutionary algorithms to define the most preferred portfolio.