The purpose was to assess the differences in spherical equivalent refractive error (SER) between cycloplegic autorefraction (1% cyclopentolate), non-cycloplegic autorefraction, and non-cycloplegic retinoscopy measured by experienced optometrists in a population with a high prevalence of hyperopia. Refractive error was measured with the three methods in 111 children aged 7–8 and 10–11 years. Bland-Altman analysis was used to assess the mean of the differences (MD) and the 95% limits of agreement (LoA) between cycloplegic autorefraction and the two non-cycloplegic methods. A mixed effects model was used to investigate the differences between methods by refractive group. Cycloplegic autorefraction gave a significantly more positive SER than both non-cycloplegic retinoscopy (MD = 0.47 D, LoA = -0.59–1.53 D) and non-cycloplegic autorefraction (MD = 0.92 D, LoA of -1.12 to 2.95 D). The mean differences in SER increased with increasing degree of hyperopia [F(4, 215) = 12.6, p < .001], both when comparing cycloplegic refraction with non-cycloplegic retinoscopy and non-cycloplegic autorefraction. Non-cycloplegic retinoscopy and autorefraction result in significantly less positive SER than cycloplegic autorefraction. The wide confidence intervals for the mean difference and limits of agreement are clinically unacceptable and the methods cannot be used interchangeably. Consequently, refraction without cycloplegia would cause misdiagnosis in some children. Even if non-cycloplegic retinoscopy results in narrower limits of agreement, the risk of misdiagnosis is not eliminated by being experienced in carrying out retinoscopy. We show that it is essential to use cycloplegia when refracting children, and in particular to ensure that no hyperope goes undetected.