Robust updating of parametric probabilistic models in the context of nonlinear structural mechanics represents a great challenge. A framework based on the combined use of structural reliability theory and Bayesian networks is proposed. The methodology is applied to practical engineering problems in the field of civil engineering. This approach appears as being useful to better estimate mechanical properties of an existing structure and may avoid carrying out in situ destructive tests. One of the major feature is that only in situ information available at the member scale (displacements, rotations, etc.) are required. Several structural cases are discussed to point out the main features of the methodology.