BACKGROUND: As more patients inquire about refractive surgical procedures, the measurement of astigmatism prior to surgery becomes more important in assessing refractive outcome. Knowledge of the repeatability of the astigmatism measurement allows one to distinguish a true change in cylinder power from measurement error. METHODS: Forty adults with structurally normal eyes and refractive errors were evaluated for the repeatability of astigmatic refractive error measures. Noncycloplegic and cycloplegic measurements of refractive astigmatism were made by retinoscopy, subjective refraction, and autorefraction. All measures were made at 2 visits within 2 weeks by the same examiner. Difference versus mean plots and the 95% limits of agreement of each technique determined the repeatability of a measurement and the agreement between the methods of measurement. RESULTS: The most reliable measure of astigmatic refractive error was cycloplegic autorefraction, with 95% limits of agreement of ±0.28 D, followed by noncycloplegic autorefraction (±0.35 D) and cycloplegic subjective refraction (±0.44 D). Noncycloplegic retinoscopy was the least reliable astigmatic refractive error measure, with interoccasion 95% limits of agreement of ±1.02 D. The most repeatable measurement of cylinder axis was cycloplegic autorefraction; none of the measurements differed by 10°or more. The least repeatable measurement was noncycloplegic retinoscopy; 40% of the measurements differed by 10°or more. CONCLUSION: For studies seeking to measure changes in astigmatism in normal eyes, cycloplegic autorefraction is the method of choice.