This paper aims at providing project office managers or portfolio-level decision-makers with an integrated methodology to guide the portfolio selection process. The issue is to define a selection algorithm which is simultaneously built on a sound theory and remains practically manageable. The originality is to integrate project interdependencies at different levels, both in the objective function (both inputs and outputs) and in the definition of new constraints that ensure the portfolio balance and effectiveness. The portfolio construction problem is formalized and resolved with Constraint Satisfaction Problem techniques. We show on a numerical fictitious example that results are significantly different considering or not project interdependencies. Some portfolios may be declared unfeasible and excluded from decision alternatives, and key values, like global revenue, may drastically change because of synergies or cannibalizations between projects. Our approach improves the performance of portfolios and increases the accuracy forecast between the expected outcomes and the actual ones, since interdependencies are modelled in a way which is closer from reality.