Objectives: To assess the accuracy of preschool vision screening in a large, ethnically diverse, urban population in South Auckland, New Zealand. Design: Retrospective longitudinal study. Methods: B4 School Check vision screening records (n=5572) were compared with hospital eye department data for children referred from screening due to impaired acuity in one or both eyes who attended a referral appointment (n=556). False positive screens were identified by comparing screening data from the eyes that failed screening with hospital data. Estimation of false negative screening rates relied on data from eyes that passed screening. Data were analysed using logistic regression modelling accounting for the high correlation between results for the two eyes of each child. Primary outcome measure: Positive predictive value of the preschool vision screening programme. Results: Screening produced high numbers of false positive referrals, resulting in poor positive predictive value (PPV=31%, 95% CI 26% to 38%). High estimated negative predictive value (NPV=92%, 95% CI 88% to 95%) suggested most children with a vision disorder were identified at screening. Relaxing the referral criteria for acuity from worse than 6/9 to worse than 6/12 improved PPV without adversely affecting NPV. Conclusions: The B4 School Check generated numerous false positive referrals and consequently had a low PPV. There is scope for reducing costs by altering the visual acuity criterion for referral.