Owing to the characteristics of the variable wind speed and small gear samples, the gear reliability of a wind turbine gearbox is hard to predict. In order to solve this problem, a complete reliability prediction model is presented in this article. Firstly, distribution parameters of the gear stress are deduced according to the variable wind data, and a linear path is defined for the time-varying stress. Next, historical gear samples are transformed into the equivalent prior data by grey relational analysis and then the posterior data is deduced by Bayes data fusion. Then, distribution parameters and a non-linear path are determined for the time-varying gear strength. After that, the dynamic reliability of gears in a wind turbine gearbox is calculated on the basis of the stress–strength interference model and the Monte Carlo sampling. Lastly, an instance is given to verify the validity of this model. The result shows that the variable wind speed will decline the reliability of wind gears, in addition, the reliability declines faster than its expectation in the infant mortality period.