In this article a reliability-based approach to determine the extreme response distribution of offshore wind turbines is presented. Based on hindcast data, the statistical description of the offshore environment is formulated. The contour lines of different return periods can be determined. Simulations are carried out for a prototype design of a 3 MW offshore wind turbine. Statistical methods are applied to determine the distribution of the extreme responses. Three approaches are used here. In the MAX approach, only the maximum of each simulation is taken into account. The POT (peak over threshold) approach takes also local maxima into consideration. The process model uses the statistical properties of the process to predict the extremes. All three methods show similar results, but POT and the process model require fewer simulations. Comparison is made for the 100 year response between these reliability-based models and a deterministic model. For this specific turbine the deterministic model underestimates the maximum flap moment but overestimates the maximum overturning moment of the support structure compared with the estimates of the reliability-based methods. The application of the reliability-based model can be extended to include other extreme load situations and achieve a more efficient structural design. Copyright © 2003 John Wiley & Sons, Ltd.