Abstract: Speckle noises are similar to intramuscular fat pixels in beef image, and they must be removed if intramuscular fat pixels need to be correctly separated from beef muscle pixels. For removing the speckle noises more quickly and completely, an improved algorithm based on adaptive fuzzy multilevel median filter method was presented. The steps are as follows: first, confirming the selected pixel is a noise or not. As for this, the absolute gap between the pixel gray value and the mean gray value of its neighbors must be calculated firstly, and then it is compared with a changeable threshold which is related to the mean gray value. If the gap is bigger than the threshold, the selected pixel will be regarded as a speckle noise. Second, if the selected pixel is a speckle noise, adaptive fuzzy multilevel median filter will be used to remove it; if not, the pixel will not be changed. The experimental results show that the improved algorithm is superior to standard median filter, multilevel median filter and morphological filter methods in preserving the details and removing the speckle noises.