This chapter deals with error quantification in estimation of models with priors of lumped linear time invariant systems comparing the conventional and set-membership identification approaches. The measured data is the only available form of truth and priors only represent the observer’s faith on the process being observed. Priors should of course be treated with skepticism. To let the data and not the observer speak of themselves and since data can do so via a model only, whatever needs to be quantified is to be modeled. Such a model is then used to compute bounds on the modeling error. In the literature on set membership identification and error quantification, the size of the posterior set of models is considered as a measure of performance of the algorithm used for quantification. However, such a measure is questionable. This chapter suggests that by considering information-rich priors on modeling errors, it is always possible (provided such priors are valid) to come up with a good performance index.