We revisit the pole-zero cancellation problem in adaptive pole-placement. It is well known that an on-line identified model in an adaptive pole-placement algorithm need not be controllable. Since the controller design usually requires that the model is controllable, this results in an ill defined control step. By exploiting the properties of the closed loop unfalsified models set, we derive an algorithm which does not suffer this difficulty The analysis is presented in the case that the system belongs to the model class.