Auto-refractors are used as a starting point for clinicians' refractions and in studies of refractive error. We investigated the repeatability of the Hoya AR-570 and clinician refraction. Eighty-six subjects, aged 11 to 60 years, were recruited by mailing inquiries to 500 randomly selected patients who had received recent examinations at the University of California Optometric Eye Center. Contact lens wearers, patients with best corrected visual acuity worse than 20/30 in either eye, and patients with a history of diabetes were excluded. Each subject was examined by two clinicians during one visit. The first clinician obtained five auto-refractor readings for each eye (which were later averaged), performed a balanced subjective refraction (with spherical masking lenses in the phoropter), and repeated the automated refractor measurements. This protocol was then repeated by the second clinician. Clinicians were randomized with regard to testing order and masked to automated refractor results, each other's refractions, and previous spectacle prescriptions. To quantify repeatability, we used mixed model analyses of variance to estimate the appropriate variance components while accounting for the correlation among, for example, repeated measurements of the same eye. Astigmatic data were analyzed by converting into Fourier form: two cross-cylinders at axis 0° (Jo) axis 45° (J45)- F o r mean spherical equivalent, the average difference between five averaged automated refractor readings, taken by two different optometrists, was +0.02 D (95% limits of agreement=—0.36 to +0.40 D). The average difference between the two optometrists' subjective refractions was —0.12 D (95% limits of agreement=-0.90 to +0.65 D). The 95% limits of agreement for the automated refractor were about half those of the clinician for both astigmatic terms (Jo and J45) and for all comparisons. Automated refraction is more repeatable than subjective refraction and therefore more appropriate for studies of myopia progression.