Abstract: For the difference of attractor between chaotic signal and noise in state space, We present a non-linear state space projection algorithm to reduce the noise of chaotic time series.Good results have been achieved on both mimic signals and practical measurement signal.The mimic signal is x-axial of Lorenz equation contaminated with 50% Gaussian white noise.The practical measurement signal is get from one single cutaneous potential recording of a pregnant mother carrying the fetus.The results indicate the effectiveness of the algorithm in reducing noise of chaotic time series.