An improved back electromotive force (EMF) harmonic adaptive compensation for PMSG sensorless control method was proposed in this paper, which is introduced into the adaptive second order generalized integrator (SOGI) based on sliding model observer (SMO). This method reduces the influence of harmonic on the rotor position in back-EMF observer and improves the precision of motor position estimation. Due to the change of the generator parameters and the nonlinearity of the converter, there is a large number of low order harmonics, such as 5th, 7th, 11th orders, in the sensorless control algorithm. In this paper adaptive SOGI is used to instead of LPF. The fundamental frequency signals is extracted by SOGI on the one hand, it reduces the low frequency harmonic content, thereby reducing the rotor position estimation error caused by low frequency harmonic, while eliminating the phase deviation caused by filtering delay. On the other hand, this method improves the system tracking performance and makes the phase error converges to zero. Finally the correctness and feasibility of the proposed method is verified by simulation and experiment.