In this paper, we utilize the observer/Kalman filter identification (OKID) and the eigensystem realization algorithm (ERA) techniques to identify the modal parameters of a centrifugal machine. To this end, we use an experimental setup to generate a pseudo-impulse input and collect output measurements which are corrupted by noise. We use the pseudo-impulse input and the OKID to find the Markov parameters of the system. Then we form the Hankel matrix of the system and determine the singular values of the system. A minimum-order, state-space model of the system is realized through the Markov parameters and then the natural frequency, damping ratio, mode shapes, and modal amplitudes at the sensor location are estimated by the ERA. We find three models for three separate cases and validate all the three identified models with the measured data and the Waterfall plot. The identified models are useful for designing passive or active vibration suppression control and fault detection systems. The results confirm that OKID/ERA is a reliable time-domain method for identifying the modal parameters of vertical centrifuge machines.