To evaluate the accuracy of autorefraction using three autorefractors comparing to subjective refraction in diagnosing refractive error in children. A cross-sectional study. setting: Community based study. study population: 117 children sampled from primary schools. procedures: All subjects underwent autorefraction using three auto refractors and subjective refraction with and without cycloplegia. main outcome measures: Spherical power, cylindrical power, and spherical equivalence (SE). Without cycloplegia, the mean SE were significantly different for Retinomax K plus 2 (−1.55 diopters, SD 2.37 diopters; 95% CI −1.98 to −1.12; P < .0001) and Canon RF10 (−1.11 diopters; SD 2.61 diopters; 95% CI −1.59 to −0.64; P = .0023) compared with monocular subjective refraction (−0.80 diopters; SD 2.25 diopters; 95% CI −1.21 to −0.35). Mean SE was significantly different for Grand Seiko WR5100K (−0.79 diopters; SD 2.40 diopters; 95% CI −1.23 to −0.35; P = .0002) compared with binocular subjective refraction (−0.62 diopters; SD 2.51 diopters; 95% CI −1.07 to −0.16). With cycloplegia, there was no significant difference in mean SE between refraction methods. Sensitivity and specificity results for the diagnosis of myopia: Without cycloplegia: Retinomax K plus 2 (sensitivity 1.0, specificity 0.51); Canon RF10 (sensitivity 0.92, specificity 0.81); and Grand Seiko WR5100K (sensitivity 0.91, specificity 0.98). With cycloplegia: Retinomax K plus 2 (sensitivity 0.97, specificity 0.99); Canon RF10 (sensitivity 0.97, specificity 0.96); and Grand Seiko WR5100K (sensitivity 1.0, specificity 0.97). Under noncycloplegic conditions, all three autorefractors have a tendency towards minus over correction in children resulting in over diagnosis of myopia. However autorefractors were accurate under cycloplegic conditions.