Proponents of the object-oriented approach advocate specialization, parameterization and reusability as the fundamental mechanisms for coping with evolution in large applications. These techniques are insufficient when dealing with incomplete specifications or ill-designed software libraries. Because prototype objects and library components need constant adjustments before they become stable classes, an iterative design approach seems appropriate. A promising methodology consists in 1) recording how standard classes must be changed when they are incorporated in new applications, 2) analyzing these modifications, and 3) restructuring the type hierarchy to suit the new requirements. This paper describes algorithms for automating this task, illustrates their characteristics through several examples, and discusses their advantages for object-oriented programming.