Abstract: Flaws in an optical fiber are automatically classified as to type by obtaining separate camera images of the fiber illuminated at different angles, and analyzing the images to detect the presence of optical patterns that are characteristic of each different type of flaw. Nominal and actual fiber boundaries are preferably first determined, preferably using noncoherent light, and an initial classification made according to the nature of the differential between the two. An analysis under a side light beam can then be performed to distinguish flaws such as splices, hard versus soft external debris, bubbles, internal debris and uncured buffer. The system reverts to a boundary light analysis to characterize a splice once it has been detected. The boundary analysis is preferably performed with a combination of diffused backlighting transmitted through the fiber, and reflected frontlighting at an angle of about 15.degree.-30.degree. to the fiber axis.