Via so-called constrained stochastic simulation, gusts can be generated, which satisfy some specified constraint. In this paper, it is used in order to generate specific wind gusts, which will lead to local maxima in the response of wind turbines. The advantage of constrained simulation is that any gust amplitude (no matter how large) can be chosen. By performing simulations for different gust amplitudes and mean wind speeds, the distribution of the response is obtained. This probabilistic method is demonstrated on the basis of a generic 1 MW stall regulated wind turbine. By considering a linearized dynamic model of the reference turbine, the proposed probabilistic method could be validated. The determined 50 year response value indeed corresponds to the theoretical value (based on the work of Rice on random noise). Next, both constrained and (conventional) unconstrained simulations have been performed for the nonlinear wind turbine model. For all wind speed bins, constrained simulation results in 50 year estimates closer to the real value. Furthermore, via constrained simulation a lower uncertainty range of the estimate is obtained. The involved computational effort for both methods is about the same.