We present a new method for identifying and segmenting the retinal pigment epithelium (RPE) in polarization sensitive optical coherence tomography (PS-OCT) images of the human retina. Contrary to previous, intensity based segmentation algorithms, our method uses an intrinsic tissue property of the RPE: its depolarizing, or polarization scrambling effect on backscattered light. Two different segmentation algorithms are presented and discussed: a simpler algorithm based on retardation data, and a more sophisticated algorithm based on local variations of the polarization state calculated from averaged Stokes vector elements. By using a state of the art spectral domain PS-OCT instrument, we demonstrate the method in healthy and diseased eyes.