We propose a logical framework for set preferences. Candidate sets are represented using profiles consisting of scalar features. This reduces set preferences to tuple preferences over set profiles. We also give a heuristic algorithm for the computation of the "best" sets.