We present a superpixel based learning framework based on retinal structure priors for glaucoma diagnosis. In digital fundus photographs, our method automatically localizes the optic cup, which is the primary image component clinically used for identifying glaucoma. This method provides three major contributions. First, it proposes processing of the fundus images at the superpixel level, which leads to features more descriptive and effective than those employed by pixel-based techniques, while yielding significant computational savings over methods based on sliding windows. Second, the classifier learning process does not rely on pre-labeled training samples, but rather the training samples are extracted from the test image itself using structural priors on relative cup and disc positions. Third, we present a classification refinement scheme that utilizes both structural priors and local context. Tested on the ORIGA−light clinical dataset comprised of 650 images, the proposed method achieves a 26.7% non-overlap ratio with manually-labeled ground-truth and a 0.081 absolute cup-to-disc ratio (CDR) error, a simple yet widely used diagnostic measure. This level of accuracy is comparable to or higher than the state-of-the-art technique , with a speedup factor of tens or hundreds.