Abstract: This paper reports tests and classification of the foreign fibers in cotton using machine vision and image processing technology in order to get numbers and weight of the foreign fiber. With gray-scale transforming and filtering technology, the authors pre-treated the original extracted image of the foreign fibers in cotton and presented image segmentation method based on self-adaptation threshold partition algorithm. The segmented binary images were processed for contour extraction and classified which takes the ratio of area of foreign fiber contour to perimeter square as the moment and adopts interior pixel points scooping method and neighborhood search method. The tests included 300 foreign fiber cotton images and the classification accuracy rate reached 96%. The results show this method is effective and accurate.