The dynamic identification of a large monumental structure is performed by using Time-Frequency Distributions. Acceleration signals are recorded on different points of the structure, allowing for the estimation of the building's parameters (i.e. frequency and instantaneous damping). The method uses auto- and cross-time-frequency transforms of the recorded signals to identify the vibration modes as a function of time and of the amplitude of the vibrations. The distributions considered have a quadratic structure, with convenient properties in the analysis of mechanical signals. In addition, quadratic distributions can use kernels in the ambiguity function domain that improve the reliability of the identification, reducing the cross terms of the distribution. The obtained results are compared with those afforded by Stochastic Subspace techniques. A model updating of the structure is also performed, starting from an indirect procedure, to provide model-based verifications.