This work proposes a system to detect visual defects in an optical fiber. Fibers of different types and with different simulated deformations were used, looking for an approximation of a real case of defect in an optical fiber. Some continuous fiber patterns were detected in images captured with a microscopic camera. The identification of these patterns was searched using different image processing techniques, such as edge detection, line detection and feature descriptors. In order to classify images of the fibers in good and defective ones, a fuzzy classifier was used. Experimental results of the algorithm are shown and is demonstrated that the proposed method helps to detect defects and classify optical fibers.