In some circumstances, it is impossible to exploit resonance crossings to identify the modal properties of rotor disks. In these cases, the identification process must rely on nonsynchronous vibrations and becomes challenging for two reasons. First, the signals are weak (compared to the levels measured during resonance crossings) and random, thus an averaging procedure is necessary. Second, the dynamical system is time-variant due to the variation of the rotor speed. This paper presents a modal identification procedure formulated in the framework of the time–frequency analysis. A region of the time–frequency plane is stretched to map the system into a fictitious linear time-invariant (LTI) system. Then, the power spectral density function (PSD) of the response is computed by an averaging procedure. Finally, the modal properties are estimated through an output-only modal identification algorithm. The procedure is applied to simulated and experimental data regarding a bladed disk of a steam turbine.