To investigate the accuracy of uncorrected visual acuity (UCVA), stereopsis, and noncycloplegic autorefraction (NCAR) tests performed by vision-screening technicians and to determine the best referral criteria when using these methods to screen for significant refractive errors in preschool children. Retrospective, case-control, and cross-sectional study. We reviewed 1000 records for a population-based preschool vision-screening program. The target conditions were defined as myopia ≤-3.0 diopters (D), hyperopia ≥ 4.5 D, astigmatism ≥ 2.0 D, and anisometropia ≥ 2.0 D. Receiver operating characteristic (ROC) curve was used to calculate optimal referral cutoff values. The examination results obtained by the vision-screening technicians were compared with those obtained by a pediatric ophthalmologist, which were considered the gold standard. The efficacies (sensitivity, specificity, positive predictive value, and negative predictive value) of different tests were evaluated. In 7.0% (95% confidence interval [CI], 5.3-8.7) of the children, at least 1 eye showed 1 of the target conditions. If only the right eyes were considered, the prevalence of target conditions was 4.2% (95% CI, 2.9-5.5). The ROC curve analysis indicated that the NCAR cylinder test (cutoff value ≥ 0.875 D) was the best test for screening target conditions. With regard to age groups, UCVA ≤ 0.75 (Snellen equivalent) and ≤ 0.85 were the best referral criteria for ages ≤ 4 years and ≥ 5 years, respectively. Combining the UCVA test with the NCAR test (the child was referred after failing both tests) increased specificity without significantly decreasing sensitivity. The UCVA and NCAR tests performed by vision-screening technicians are adequately sensitive and specific for preschool vision screening. The ROC curve analysis was used for determining the appropriate screening criteria for these tests, and combining the tests increased their accuracy. The screening criteria should be age dependent. When analyzing the test accuracy in ophthalmic problems, if the disease of interest does not symmetrically (in terms of disease severity and prevalence) involve both eyes, the prevalence based on only 1 eye should be interpreted with caution.