We describe a context-sensitive model of choice, in which the selection process is shaped not only by the attractiveness of items but also by their semantics ('salience'). All items are ranked according to a relation of salience, and a linear order is associated to each item. The selection of a single element from a menu is justified by one of the linear orders indexed by the most salient items in the menu. The general model provides a structured explanation for any observed behavior, and allows us to to model the 'moodiness' of a decision maker, which is typical of choices requiring as many distinct rationales as items. Asymptotically all choices are moody. We single out a model of linear salience, in which the salience order is transitive and complete, and characterize it by a behavioral property, called WARP(S). Choices rationalizable by linear salience can only exhibit non-conflicting violations of WARP. We also provide numerical estimates, which show the high selectivity of this testable model of bounded rationality.