This paper accurately quantifies the way in which noise induced estimation errors are dependent on model structure, underlying system frequency response, measurement noise and input excitation. This exposes several new principles. In particular, it is shown here that when employing Output--Error model structures in a prediction-error framework, then the ensuing estimate variability in the frequency domain depends on the underlying system pole positions. As well, it is also established that the variability is affected by the choice of model structure, in that it is twice as much when system poles are estimated as when they are a-priori known and fixed, even though the model order is the same in both cases. These results are unexpected according to pre-existing theory.