Since the fault diagnosis of wind direction sensors is extremely concealed owning to its Gray code encoding, traditional fault detection methods need to disassemble the wind direction sensor, which destroys the integrity of meteorological observation data. A measurement model for wind direction sensors based on probability distribution is built, based on which, a fault detection algorithm in terms of distribution law rules is proposed. A graphic wind direction sensor fault detection program is developed though VB/SQL/MATLAB hybrid programming technology. Finally, the fault detection test is carried out, and the results show that the fault detection algorithm can recognize the single or mixed faults exactly without disassembling the wind direction sensor.