We study several situations where it is possible to extract a feasible computational information from a proof, either in a first-order theory or in a propositional proof system.