Abstract: The Fuzzy C-means algorithm has been successfully used for clustering of data in pattern recognition for a number of years. In this paper, a new noise smoothing scheme using the Fuzzy C-means algorithm is proposed. In general, a pixel is noise if the difference between it and its surrounding ones is greater than a given threshold. Based on this fact, first, we use Fuzzy C-means algorithm to cluster pixels in a 3 by 3 mask, then use a standard nucleus function (a window function) to find out noises in this area, and at last, replace the noise by the average of its surrounding area. Due to the advantage of this smoothing scheme that only noises are to be changed, the others remain unchanged, noises can be cleaned out very well and the edges and details well reserved. Nine pixels are processed simultaneously instead of only one every time. So the computation speed is fast. Results of the application of this scheme to several real world images, the evaluation of its performance, and a comparison between it and the Gradient Inverse Weighted Smoothing scheme are also presented.