Bearing dynamic characteristics have been a major unknown in the modelling and analysis of large turbo-generators. An identification algorithm for bearing dynamic characterization by using unbalance response measurements is developed for multi-degree-of-freedom (m.d.o.f.) flexible rotor-bearing systems. The algorithm identifies the bearing dynamic parameters, consisting of four effective stiffness and four damping coefficients for each bearing, utilizing frequency domain synchronous unbalance response measurements from the accelerometers attached to the bearing housings in the horizontal and vertical directions, for a minimum two different unbalance configurations. The procedure of identifying bearing dynamic coefficients by using the proposed algorithm is presented and demonstrated through a numerical example. Adding noise to the simulated signal checks the robustness of the algorithm against measurement noise. Combinations of regularization and the generalized singular value decomposition (SVD) are used to tackle an ill-posed problem due to the nearly circular orbit of the rotor at the bearings, as a special case for nearly isotropic bearings. It is demonstrated that by measuring noisy bearing responses with the direction of rotation of the rotor both in the clockwise and anticlockwise directions, the bearing estimation problem for circular orbit becomes well-conditioned. The regularization algorithm is tested for an experimental rotor-bearing rig. The response reproduction capabilities are excellent even in the presence of measurement noise.