The European Academy of Wind Energy has highlighted main bearing failures as a significant concern for enhancing the reliability and availability of wind turbines, due to the extensive repairs, substantial replacement costs, and prolonged downtimes these failures cause. Consequently, main bearing fault prognosis has emerged as both an economically significant and technically challenging area. This paper introduces a data-driven methodology for fault prognosis. The primary contributions of this study are outlined as follows: Prognosis is achieved through the utilization of SCADA (supervisory control and data acquisition) data, which is readily available in all industrial-sized wind turbines, eliminating the need for additional purpose-specific sensors. The proposed methodology (based on training a normal behavior model) requires only the collection of healthy data, allowing for application across any wind farm, even in the absence of recorded faulty data. Additionally, the proposed algorithm is effective under varying operational and environmental conditions. The effectiveness and performance of this methodology are validated on an operational wind farm comprising 12 turbines. Results demonstrate that advanced prognostic systems, which rely solely on SCADA data, can predict failures several months in advance, enabling wind turbine operators to strategically plan their operations.