In this paper, we propose a new image processing method to recognize the fiber patterns in the image of PET/Rayon composite yarn cross-section. Our method consists of two voting techniques: the connected component voting (for obtaining single fiber locations) and the circle parameter voting (circle detection, for recognizing the fiber patterns). When compared with the previous approach, the new method needs fewer parameters and is more flexible. Several real images are used to show the performance of the proposed method.