Aiming at the problem of high noise in the output signal of MEMS gyroscopes, which affects measurement accuracy, a denoising method for MEMS gyroscopes based on an improved wavelet transform threshold selection strategy is proposed by combining signal correlation and distance based classification methods. This method first utilizes the correlation between signal components after wavelet decomposition, constructs a correlation function, and distinguishes the noise content of different components. Then, based on Block distance, the wavelet coefficients of each signal component are classified. Combining the number of wavelet decomposition layers and the value of the correlation function, the wavelet coefficients are screened to remove noisy wavelet coefficients, and the wavelet coefficients of the signal components after wavelet decomposition are classified and reconstructed. The experimental results show that under dynamic signal conditions, compared with the traditional threshold selection strategy, the improved strategy improves the signal-to-noise ratio of the signal by 7.073 dB and increases the mean square error by an order of magnitude.