Abstract. Model differences calculated by differencing algorithms contain the atomic changes made to a model. However, they do not capture the user’s intention of the modification. We present concepts and a frameabstracting from atomic changes to produce semantic changes, for example, “move all classes from package A to B” instead of “move X, Y, and Z from package A to B”. Semantic changes abstracted this way are closer to the user’s intention and are applicable to other models much like customizable refactorings.