Various tasks need to consider preferences in a dynamic way. We start by discussing several possible meanings of preference change, and then focus on the one we think is the most natural: preferences evolving after some new fact has been learned. We define a family of such preference change operators, parameterized by a revision function on epistemic states and a semantics for interpreting preferences over formulas. We list some natural properties that this kind of preference change should fulfill and give conditions on the revision function and the semantics of preference for each of these properties to hold.