Abstract: A framework for belief change is introduced in which the selection mechanism operates directly on potential outcomes rather than on possible worlds or maximal consistent subsets of the original belief set. Success conditions are expressed with (metalinguistic) belief descriptors that are built from atomic sentences of the form Bp where p is a sentence in the object language. In this way a general form of belief revision is obtained that includes revision and contraction as special cases. Revision by a sentence p is expressed with the success condition {Bp}, multiple revision has success conditions of the form {Bp1,  Bp2…}, contraction has {-Bp}, etc. New operations can also be defined, such as “making up one’s mind” {Bp V B-p}. A distance-based semantic model for descriptor revision is proposed in which the outcome is the closest potential outcome in which the success condition is satisfied. The outcome set (set of belief sets that are potential outcomes of belief change) is taken as given. A generalization of epistemic entrenchment, epistemic accessibility, is defined that applies to all success conditions. It has standard entrenchment (for contraction) as a special case. The new approach is motivated by problems in previous models (such as the recovery property) that are avoided in the new framework.