This paper presents a technique for using a parametric approach for modal properties estimation of a floating offshore structure. The technique utilizes empirical mode decomposition (EMD)-based time-varying autoregressive (TVAR) model, which is an extended form of the well-known AR model. Input for TVAR model is surge motion as system response, obtained experimentally from a scaled 1:40 model of a prototype semi-submersible. By utilizing the advantage of time-varying spectrum generated from TVAR model, estimation of the modal properties is carried out in the time-frequency plane via poles technique under different random wave spectrum. The results show that the modal frequency and its corresponding damping ratio for surge motion either in low frequency or wave frequency region can be estimated well. The Stochastic subspace identification method and free-decay test of the semi-submersible prototype are taken as benchmark.