The random shift error of MEMS gyroscope is the main error of this kind of sensor and the conventional de noising method with the hard and soft threshold has some shortages in the de noising effect and the real time ability. In this work, the de noising principle of the lifting wavelet threshold has been analyzed based on the basic theory of the lifting wavelet transform and an improved threshold de noising method based on the lifting wavelet has been proposed. Combining with MEMS gyroscope output data, the simulation experiments of the de noising method with hard and soft threshold and the improved threshold de noising method have been compared and analyzed. The results showed that the higher signal noise ratio of 83.5327 dB has been obtained by using the improved threshold de noising method. Moreover, the method proposed in this work has better de noising effect and faster process speed comparing with the conventional de noising method with the hard and soft threshold, thus is more suitable for the applications.