Parapapillary atrophy (PPA) is a crescent atrophic abnormality adjacent to optic disc (OD), which lies in retinal pigment epithelium, photoreceptor and laminae choriocapillaris. Clinically, PPA is an indicator for clinical or computer aided diagnosis of pathologies of eyes, such as myopia and glaucoma. In this paper, an algorithm is proposed in order to automatically segment PPA from retinal images. Firstly, evenly-oriented radial line segments were used to detect the candidate boundary points of OD and augmented-OD (A-OD, which indicates the region that combines PPA and OD). Then, to obtain smooth, non-discrete boundaries of A-OD and OD, those points are fitted to ellipses by the least square method. Finally, the boundary of PPA is obtained after finding the non-overlapping part of these two boundaries. The algorithm needs the coordinate of the center of OD and its approximate radius as inputs, with assurance that input retinal image contains PPA. Experimental results based on 100 retinal images tested show that the average F-score of this algorithm reaches 0.67.