In the assessment of food quality grade, the number of black germ kernels will affect the overall classification of wheat. With 710 wheat kernels containing black germ kernels, 627 break kernels and 1 167 normal wheat kernels produced in Xinjiang and Henan province，the color linear CCD was used to acquire images of the wheat kernels. After removing the margin of original image background, correcting color offset, image enhancing, morphological processing and ternary of the original image, the feature of black germ kernels was extracted effectively. In the experiment, the recognition accuracy of black germ kernel, break kernel and normal kernel were 95.1%, 96.0% and 98.3%, respectively. The results show that the method is feasible and effective to segment black germ kernels.