When discussing a feedback control system, we must aware that no mathematical model can exactly model a physical system. We concern about how modelling errors might adversely affect the performance of a control system. Suppose there is a model p belonging to a set P, any other model in set P can be viewed as the model p plus a kind of modelling error. A feedback control system is called robust with respect to stability if this feedback can make every model in P stable. Obviously, the notion of robustness is only valid for a certain model set. If a feedback control system is stable, we cannot say that this feedback is always robust for any kinds of modelling uncertainty.