This paper evaluates the repeatability of refraction in keratoconus patients and normal myopes, using representations of spherocylindrical power that are theoretically valid. Data consist of test-retest refraction data from the Collaborative Longitudinal Evaluation of Keratoconus (CLEK) Study and similar data from normal myopes. Dioptric powers are transformed to an orthogonal vector representation of dioptric power. The metric of change is the dioptric difference between test and retest. The median difference between test and retest in keratoconus patients is four to six times larger than in normal myopes. Refraction over a rigid contact lens on a keratoconic cornea improves repeatability but remains approximately twice as large as in normal myopes. The methods demonstrated here possess advantages over previous methods and provide a more valid comparison between test and retest and between different subject groups. The repeatability of refraction in keratoconus patients is substantially worse than in normal myopes.