Given a concept hierarchy and a set of instances of multiple concepts, we consider the revision problem that the primary concepts subsuming the instances are judged inadequate by a user. The basic strategy to resolve this conict is to utilize the information the hierarchy involves in order to classify the instance set and to form a set of several intermediate concepts. We refer to the strategy of this kind as hierarchy-guided classification. For this purpose, we make a condition, Similarity Independence Condition, that checks similarities between the hierarchy and the instances so that the similarities are invariant even when we generalize those instances to some concept at the middle. Based on the condition, we present an algorithm for classifying instances and for modifying the concept hierarchy.