The invention discloses a method for detecting the fiber distribution in a mixed fiber product. The method comprises the following steps: 1, collecting an original image of the mixed fiber product, and obtaining a low-noise gray scale image through denoising and gray scale transformation; 2, setting any fiber as target or non-target fibers; 3, cutting that image of the step '1' into small pixel maps, and selecting an appropriate threshold value to process the small pixel maps; 4, extracting that eigenvalue of the non-target fiber as noise, performing two-dimensional multi-level stochastic resonance on the image of the step '3', and extracting the target fiber accord to the result; 5, binarizing that image of the step '4', recombine the small images into the binarized images of the original size accord to the original cutting order, if the binarized images can display the target fiber outline, carrying out the next step, otherwise returning to the step '4'; 6, counting that number of target fib pixels of the restored image, and giving the fib distribution value of the mixed fiber product. By extracting the eigenvalues of the target and non-target fibers and by removing the non-target fibers to highlight the target fibers, the present invention has the advantages of simple operation and high processing accuracy.