The evaluation of the seismic safety of existing structures is strongly affected by the incomplete knowledge of the mechanical parameters of the materials and constructive details, after the diagnostic investigations and the technological and geometric survey. The mechanical parameters influence the response in a random way, not only because of the uncertainties related to the limited reliability of the in-situ tests, but also because of the intra-building variability of these parameters. In the case of constructive details, incomplete knowledge may in some cases condition the choice of model, thus introducing epistemic uncertainties.

At the standard level, depending on the knowledge level achieved, the problem is faced through an approach that leads to the definition of confidence factors, to be considered in the verification procedure with the aim of penalising it, i.e. leading to a more cautionary assessment when knowledge is limited. However, the way these factors are determined and applied is largely arbitrary. Indeed, these factors are established a priori, without really taking into account the propagation of uncertainties on the response, and are applied arbitrarily to the material strength parameters, which are not always the ones that most influence the outcome of the verification.

The work will critically examine the current standard approaches and apply them to a number of prototype buildings, for which the possible uncertainties compatible with the different knowledge levels, according to the standards definition, are considered. In particular, nonlinear static analyses are carried out to quantify the propagation of uncertainties, by using a Monte Carlo sampling (100 models), distinguishing between those deriving from the mechanical parameters or masses and those associated with drift. It is worth noting that the drift uncertainties are in most of the cases the dominant ones, and cannot be reduced by investigations.

Having verified the unreliability of the current approach, a practice-oriented procedure will be proposed in the paper, based on a codified sensitivity analysis, which allows to derive the confidence factor from the dispersion of the outcome of the assessment, which is the input acceleration compatible with the limit state under verification. Therefore, the proposed procedure is consistent with a probabilistic safety check, even if based only on few deterministic evaluations.