In the present paper methods for statistical load extrapolation of wind turbine response are studied using a stationary Gaussian process model which has approximately the same spectral properties as the response for the flap bending moment of a wind turbine blade. For a Gaussian process an approximate analytical solution for the distribution of the peaks is given by Rice. In the present paper three different methods for statistical load extrapolation are compared with the analytical solution for one mean wind speed. The methods considered are global maxima, block maxima and the peak over threshold method with two different threshold values. The comparisons show that the goodness of fit for the local distribution has a significant influence on the results, but the peak over threshold method with a threshold value on the mean plus 1.4 standard deviations generally gives the best results. By considering Gaussian processes for twelve mean wind speeds the ‘fitting before aggregation’ and ‘aggregation before fitting’ approaches are studied. The results show that the ‘fitting before aggregation’ approach gives the best results.