For the shortcoming of traditional wavelet neural network(WNN), a WNN algorithm based on modified unscented
Kalman filter(UKF) is proposed. The algorithm uses an UKF based on Sigma point of simplex spherical distribution(SSUKF) to train the parameters of WNN, which can improve the learning ability and training quality of WNN. The experiment results on aerodynamic modeling of flight vehicle show that, compared with BP and extended Kalman filter(EKF), the WNN trained by SSUKF algorithm has a better ability with features of convergence, precision and calculation, and is also a good method for aerodynamic modeling of flight vehicle.