A novel system identification technique, Seismic Frequency Domain Decomposition (SFDD), has been developed and validated to estimate the dynamic properties of structures from Multiple-Input–Multiple-Output (MIMO) seismic data. SFDD extends classical FDD and CMIF approaches by incorporating seismic input data and using smoothed Frequency Response Function (FRF) matrices obtained via Welch’s method or ARX-based estimation. From the Singular Value Decomposition (SVD) of the equivalent CMIF matrix, natural frequencies and mode shapes are extracted, while damping ratios are computed using broadband methods. Additionally, the SVD of Power Spectral Density (PSD) matrices, assuming individual seismic inputs, provides a new frequency representation describing the seismic participation of each input. This enables the estimation of seismic-normalized mode shapes and influence vectors, as well as disaggregated transfer functions for each modal response. The method is validated against a finite-element model of the JEAM building and applied to real seismic data from a low-intensity earthquake, yielding results comparable to those of the DSI method. A time-frequency version, Short-Time SFDD (STSFDD), allows tracking time-varying properties and modal responses. STSFDD is successfully applied to the 2010 Chilean mega-earthquake (Mw=8.8), revealing the nonlinear seismic behavior of the JEAM building.