A method of amplitude frequency characteristics compensation is presented to realize ultra-low frequency vibration measurement based on wavelet neural network(WNN) for vibration velocity transducer. In this method, a dynamic compensation network can be set up according to measurement data of dynamic response of vibration velocity transducer. The compensation principle is introduced and the geometrical structure of the network is analyzed and the algorithms for the training and initialization of network parameters are given. The weights of network, scale factor and displacement factor are trained by the steepest descent method and the network parameters initialization is integrated with the wavelet type, time-frequency parameters of wavelet and the training samples. The results show that the proposed wavelet neural network has good robustness, on-line correction ability, and higher precision and faster training speed than the BP neural network when used in the amplitude frequency characteristics compensation of vibration velocity transducer, and has practical value in measurement field.