This paper presents a novel approach for online operational modal decomposition utilizing nonstationary noisy measurements only. Critical problems include the online updating of modal participation factors and the estimation of mode shape components and they are successfully resolved in the proposed online operational modal decomposition framework. The algorithm is performed in a two-phase manner. In the first phase, mode shape components for the observed degrees of freedom are extracted from the nonstationary noisy measurements. In the second phase, modal coordinates are updated recursively by combining a modal extended Kalman filter and a Bayesian probabilistic approach for online updating the modal participation factors. The proposed algorithm requires neither a finite element model of the dynamical system nor the assumptions about stationarity of the measurements and, thus, it provides a computationally efficient framework for online operational modal decomposition of dynamical systems. An illustrative example and a practical application are utilized to demonstrate the applicability and efficiency of the proposed approach.