The traditional low- order adaptive filter is inefficient in suppressing or eliminating the higher- order statis-tical noises of laser displacement signals,and the convergence speed as well as the SNR (Signal- to- Noise Ratio) ofblind source separation based on the maximum entropy or the natural gradient may remarkably decrease under mi-cro- displacement measurement conditions.In order to solve these problems,a maximum 3- shifted correlated kurto-sis deconvolution (M3CKD) algorithm is presented on the basis of the kurtosis variation of mixed signals,and thecorresponding inverse filter is designed.Moreover,the convergence and stability conditions of the algorithm areanalyzed,and a blind extraction is carried out on a constructed experiment platform.The results indicate that theproposed algorithm is efficient in the blind extraction of laser displacement signals and multiple reflection signals,and that it outperforms the FastICA algorithm due to its high convergence speed and large signal- to- noise ratio.