Wind farms must periodically take their turbines offline in order to perform scheduled maintenance repairs. Since this interruption impacts the generation of energy and any shortfall in production must be covered by energy purchases in the spot market, determining the optimal time to start maintenance work at a wind farm is key to maximizing your revenue, which is a function of both the expected wind speeds and electricity spot prices. In this study we develop a model to determine the optimal maintenance schedule in a wind farm. We analyze a window of opportunity in the most likely period of the year and perform weekly updates of expected wind speeds and energy price forecasts. Wind speeds are forecasted with an ARIMA model, while spot prices are simulated under the Newave dual stochastic programing model. The decision to defer maintenance to a future date is modeled as an American real option. We test two models with actual data from a wind farm in the Brazilian Northeast, and compare our results with current practice and with maintenance scheduling considering perfect information in order to determine the benefits of the model. The results suggest that the models may provide significant advantages over a stopping decision that randomly chooses a week to begin maintenance within the opportunity window and is close to the ideal optimal stopping date considering perfect model.