Several nonlinear system identification methods have been presented in the past, such as the Extended Kalman Filter, the H filter and the Sequential Monte Carlo methods. One of the most promising ones is the Unscented Kalman Filter (UKF) recently proposed for the on-line identification of structural parameters. In the present study the UKF is proposed to the purpose of the nonlinear identification of a new prototype of rolling-pendulum tuned vibration absorber which, relying on an optimal three-dimensional guiding receptacle, can simultaneously control the response of the supporting structure along two orthogonal horizontal directions. Unlike existing ball vibration absorbers, mounted on spherical recesses and used in axial-symmetrical structures, the new device can be bidirectionally tuned to both fundamental structural modes even when the corresponding natural frequencies are different, by virtue of the optimum shape of the rolling cavity. Based on preliminary numerical simulations, the UKF is shown to be effective in identifying the structural parameters of the new device and particularly the nonlinear rolling friction dissipation mechanism at the interface between the ball bearing and the rolling surface.