This paper deals with model validation in structural dynamics for a family of quasi-identical structures in the context of uncertain measurements. The crucial point is for the engineer to be able to quantify the quality of the model, which is probabilistic with respect to a set of measurements from which a probability density function can be extracted. Our approach is based on the ”mechanical concept” of Constitutive Relation Error Estimator (CRE), which was introduced initially in order to quantify the quality of finite element analyses, then developed in the deterministic context. Our extended CRE estimator enables us to quantify the quality of a given probabilistic model and, thus, to update and validate the model. Several examples are given, including an industrial case.