A nonlinear filtering algorithm was presented to remove salt and pepper noise from gray-scale image while preserving details.Utilizing the statistical information in a local window of the image,each pixel is classified roughly to be signal pixel,possible positive noise pixel and possible negative noise pixel,and the noise labeling matrix is built.Then the possible noise pixels are subdivided into signal pixels,noise pixels and uncertain pixels according to the statistical information of the noise labeling matrix.Different filtering schemes are employed for different kind of pixels to preserve the image details.Comparing experiments are also done for five different filtering algorithms include the algorithm presented in this paper.The simulation results show that our algorithm outperforms the other four algorithms in the aspects of filtering ability,adaptability and details preserving ability,especially for highly corrupted cases.