In the adaptive optics system, the piezoelectric steering mirror(tip/tilt mirror, TTM) is usually used to correct the wavefront aberration caused by atmospheric turbulence in real time. However, the response of the piezoelectric tilting mirror has large nonlinear hysteresis effect, which greatly reduces the precision of the tilting mirror in place, affects the stability of the system, and restricts the bandwidth of the skew correction system. Therefore, the hysteresis phenomenon needs to be modeled and compensated by the established model. In this paper, hysteresis operator is introduced and using Bayesian regularization training algorithm to train BP (back propagation) neural network to construct hysteresis model of piezoelectric steering mirror. Then experimental study was conducted on a piezoelectric steering mirror developed by Institute of Optics and Electronics, Chinese Academy of Sciences. The final experimental results show that the hysteresis model of piezoelectric steering mirror constructed by BP neural network has more accurate identification capability, the hysteresis size in the X direction decreased from 6.5% to 1.3% and that in the Y direction decreased from 7.1% to 1.6%.