This work presents a new approach for nonlinear system identification of floating structures using time-varying autoregressive with exogenous input (TVARX)-based Volterra model. This method utilizes time series of measured wave heights as system input and surge motion as system output. Solution to the proposed approach is also proposed via Kalman smoother algorithm. The efficacy of the proposed method is then applied in a scaled 1:100 model of a prototype truss spar. It is shown that surge motion obtained from the identification results has good agreement with the experimental results either in time or frequency domain. © (2014) Trans Tech Publications, Switzerland.