A method and device for predicting a wind turbine fault is provided. The method includes: initializing a wind turbine model of the wind turbine; updating wind turbine model parameters of the wind turbine model based on a current environment condition and current actual wind turbine state parameters of the wind turbine periodically; establishing a variation model of the wind turbine model parameter based on historical wind turbine model parameters and corresponding historical environment conditions; predicting the wind turbine model parameters at a future moment by using the variation model of the wind turbine model parameters based on a future environment condition at the future moment; establishing a wind turbine fault model based on the historical wind turbine model parameters and corresponding historical wind turbine faults; predicting the wind turbine fault at the future moment based on the predicted wind turbine model parameters and the wind turbine fault model.