In this paper, the impact of several sources of uncertainty on the seismic response of an experimentally tested building is assessed for the case of a post-earthquake scenario. The building comprises unreinforced masonry and reinforced concrete walls. It is modeled using an equivalent frame model and latin hypercube sampling is used to generate randomized nonlinear models of the building. A sensitivity analysis is performed in order to identify the model parameters that contribute the most to the response. To deal with several sources of uncertainty and take advantage of the information of the building at the end of the fifth ground motion sequence of the experimental campaign, a data-interpretation methodology called error-domain model falsification is implemented. This methodology involves discarding models that do not reasonably match the measured response in terms of post-earthquake frequency and damage grade, which is determined from the maximum roof displacement. The models that are not falsified are used to predict the response of the building for the final four runs, which is assessed here in terms of the maximum roof displacements. The falsification procedure significantly reduces the number of models that reasonably explain the observed behavior and improves the reliability of response predictions for all ground motions following falsification. Results are accurate in the sense that the experimentally obtained response lies always within the range of predicted responses.