The Analog Ensemble (AnEn) has already been applied to generate probabilistic forecasts of weather variables, renewable energy, and electricity demand. It uses a historical set of forecasts from a weather model or other forecasting systems and measurements of the variable to be forecast. For each forecast horizon, the ensemble consists of a set of past observations coinciding with past forecasts most similar to current forecasts for the same time frame. Recent applications have shown that the AnEn introduces a negative bias when predicting events in the tail of the predicted wind speed distribution, particularly when the training dataset is short. This underestimate increases when the predicted event occurs less frequently in the historical data. Therefore, a new bias correction for the AnEn was tested using wind observations from more than 500 US stations to reduce the underestimation associated with rare events. It was shown that the negative bias introduced by the AnEn is significantly reduced by our new approach. In addition, the overall probabilistic performance of AnEn improves when wind speeds above 10 m/s are expected, as demonstrated by the continuous ranked probability score values. These improvements can be attributed to improved reliability obtained by introducing the proposed correction algorithm.