A method based on response-conditioned waves is presented, for the probabilistic assessment of the short-term extreme responses of offshore structures. It is demonstrated that this method has the potential to be an efficient alternative to Monte Carlo simulations, which might be very time-consuming when low probability events are sought. The key element of the proposed approach is the use of the nonlinear wave propagation model HOS-NWT, which allows for the direct and accurate reproduction of those wave episodes in experiments or with high-fidelity numerical simulations. Under these terms, the equivalent fully nonlinear response at a given probability level is obtained and a discrete response distribution can be constructed. The method is experimentally validated for the prediction of the nacelle acceleration of a SPAR-type floating offshore wind turbine.