The existing cross-entropy thresholding segmentation methods can not segment the image with mixed noise efficiently. The 3D cross-entropy thresholding method and its recursion formula are proposed based on the 3D histogram of an image. The experiment results show that the 3D thresholding method can get better segmentation results for the images with the mixed noise than the existing algorithms due to considering the gray value, the local average gray value and the neighbor median gray value of the image.