The invention discloses a variable coefficient torque control based wind machine's maximum power point tracking control method. Based on the foundation of a traditional power curve method and in a torque adjustment, an optimal torque is multiplied by an adjustable gain coefficient so as to obtain a higher wind energy capturing efficiency. The gain coefficient is adjusted in this way that based on the relationship between a best gain coefficient and the statistics of the wind characteristics, a fitting expression could be obtained, and the fitting expression is used to periodically adjust the gain coefficient. According to the method, the gain coefficient can stay or near to the best gain coefficient even when wind speed changes. Consequently, the obtained gain coefficient can better reflect wind conditions and the distributional characteristics of wind energy so that the tracking performance of a wind machine can improve along with wind conditions and obtain the greatest possible wind energy capturing efficiency. Further, in the coefficient adjustment, all one needs to do is measure the turbulent flow intensity of a sampled wind speed in an iteration period. With no further information, the method can be very convenient to perform.