Abstract: The noise density of Electron Multiplying CCD(EMCCD) image varies with the gain, a noise detection based on adaptive fuzzy median filter(AFMF) algorithm was proposed. The algorithm consisted of fuzzy filtering module and adaptive module. First, the noise pixels in the center of the filter window was identified. Second, the double thresholds were introduced for these detected noise points, basing on the thresholds and median of the filtering window, the fuzzy membership function of noise points was put forward, and the fuzzy membership function was utilized to filter the noise points. Finally, the adaptive module was used to adjust the pixel in the image. Simulation and experimental results indicate that the new algorithm is able to remove noise pixels effectively and protect the details well in the image. Compared with the adaptive median filtering, the average PSNR improves at least 15 dB. The performance is better than the other median filters under the condition of low noise density and relatively stable under the condition of high noise density.