In this paper, a new theory of qualitative comparative descriptions for dynamic system behavior is presented. System deviations and behavior deviations are viewed relative to the normal case. In contrast to existing approaches, a deviation is not only characterized as "less than normal" or "greater than normal" (LGTN), but deviations can also be compared with each other in order to avoid ambiguities and provide more precise predictions. A fundamental problem in comparative behavior prediction is that LGTN deviations can cause non-LGTN effects like a change of the direction of a parameter or a change in the order of events. Such so-called changes in the behavioral topology (10,ll) cannot be handled by existing approaches in a satisfying way, but are covered by our theory. Our theory is incorporated into the relative simulator RSIM+. RSIM+ can be viewed as an extension of the QSIM simulator (7). It provides a refined system description with qualitative predictions which have not been achieved in other work. In particular, it is guaranteed that all behaviors following from an LGTN deviation are predicted.