Pre-processing of the gyro signal is an effective way to improve the accuracy of the gyro. In this paper, the frequency band analysis method based on wavelet packet is used to detect the useful signal from noise which was measured at different angular velocity inputs, thus the effective frequency range gyro signal is identified. Then the gyro's output is compensated for by wavelet packet dynamic threshold method at different sub-frequency range, thus the smooth motion characteristics of gyro could be improved, and the gyro has high precision in the overall measurement range. The results based on practical measured data show that this method can be applied in the scope of full-scale gyro error compensation, and can effectively improve the accuracy of the gyro output signal.