In order to warn the potential failure of wind turbines and enhance the safety of the unit output, a wind turbine fault warning method with abnormal data reconstruction is proposed based on the supervisory control and data acquisition (SCADA) system. Firstly, the SCADA data of the wind turbines in the same wind farm are fully utilized to reconstruct the two types of target unit data of the input and output, respectively, which overcomes the problems of partial data missing and data anomaly. Secondly, a fault warning model is established using the extracted representative data, which is closer to describe the dynamic characteristics of the unit. Thirdly, the improved deterioration degree is adopted to warn the potential failures and intuitively show the phased deterioration of the unit. In the case study, the SCADA fault data of a certain wind farm were used, and the parameter settings of the proposed strategy were determined with three criteria. The results show that the proposed method can predict the potential fault of the gearbox of the wind turbine at least 3 weeks ahead, which verifies the timeliness of the proposed early warning method.