Representing (and reasoning about) preference relations over combinatorial domains is computationally expensive. For many problems involving such preferences, it is relevant to simplify them by projecting them on a subset of variables. We investigate several possible definitions, focusing without loss of generality on propositional (binary) variables.