Identification of transfer functions from surge motion response of a semisubmersible platform from model test data is presented in this paper. The identification is carried out by estimating the time-varying model coefficients of a time-varying NARX (TVNARX) model. The coefficients are estimated using proposed method, named artificial bee colony-based Kalman smoother (ABC-KS). System input-output data for identification process are wave height and surge motion from a scaled 1:100 model of a prototype semisubmersible. The applicability of proposed method is assessed numerically and experimentally under unidirectional long-crested random waves. The results show that the linear and quadratic frequency response functions as well as the wave and low frequency responses of a semisubmersible platform can be well identified either in time or frequency domains. The evolution of the nonlinear wave-structure interactions also can be revealed with respect to measured time duration. © 2015 Elsevier Ltd.