Purpose: Compare the performance of different amblyopia screening tests. Methods: Based on exploratory factor analyses (EFA) of different screening tests performed in 3295 children, we created models of screening strategies in a matrix with: uncorrected visual acuity (UCVA), Plusoptix measurements (PO), Randot Stereo-test (SR), and Cover-Test (CT). Receiver Operating Characteristic (ROC) curves and confusion matrix were used to compare performance of different model’s algorithm to predict new diagnosis of amblyopia. Estimated screening costs per screened and treated child were compared. Results: Regression analyses revealed that, although all models predicted amblyopia (all p < 0.001), only models including PO or UCVA had higher prediction capacity ( R2 > 0.4) and better discriminating ROC curves (AUC > 0.95; p < 0.001). For 96% sensitivity, UCVA + PO was the most cost-effective model, since the estimated average screening costs per treated child, almost doubled and tripled if using PO or UCVA alone, respectively, versus using both exams. When UCVA + PO is not possible to implement, adding SR to either UCVA or PO resulted in cost-savings of 28% and 18%, respectively. Conclusions: In a previous unscreened population, aged 3–4 years, screening programs using either UCVA or PO alone, should reconsider doing both tests simultaneously, since, for a high level of sensitivity, using simultaneously UCVA + PO is more cost-effective, per screened, and treated amblyopia. Concerns relating higher time-consuming exams for the combination of UCVA + PO should be surpassed, since costs per treated child drop considerably. When children benefit from good primary-care routine examinations since birth, no benefit was found for using CT in a screening setting. SR showed little benefit.