In this paper, a novel scheme for subsynchronous oscillation detection and modal parameter estimation is proposed, by leveraging the rich information contained in high-rate phasor measurements as well as the effectiveness of synchrosqueezing transform for multimodal signal analysis. Specifically, an instantaneous time-frequency representation of a voltage/current signal is first obtained by applying synchrosqueezing transform to the real-time data collected by a phasor measurement unit. The non-zero synchrosqueezing transform coefficients quantify the undamped frequency components of the original voltage/current signal at each time instant. For an unknown number of undamped frequency components, unsupervised clustering is applied to the non-zero synchrosqueezing transform coefficients in the frequency domain, so as to determine how many modes comprise the signal, as well as which mode each non-zero synchrosqueezing transform coefficient belongs to. Then, for each detected mode, the corresponding non-zero synchrosqueezing transform coefficients are utilized to reconstruct a component of the original voltage/current signal. Finally, the magnitude, damping factor and phase angle of each mode are estimated by applying a least square estimation algorithm to the reconstructed component signal. The effectiveness of the proposed approach is revealed through several case studies using IEEE benchmark models. Further, practical issues involving missing data, measurement noise and transform basis functions are also systematically addressed in this study.