The Eating Assessment Tool-10 (EAT-10) is used worldwide to screen people quickly and easily at high risk for swallowing disorders. However, the best EAT-10 cutoff value is still controversial. In this systematic review and meta-analysis, we estimated and compared the diagnostic accuracy of EAT-10 cutoff values of 2 and 3 for screening dysphagia. We searched the PubMed, Web of Science, EMBASE, Cochrane Library, CNKI, WANFANG, and VIP databases from May 2008 to March 2022. The meta-analysis included 7 studies involving 1064 subjects from 7 different countries. Two studies were classified as high quality and five studies as medium quality. With an EAT-10 cutoff value of 2, using flexible endoscopic evaluation of swallowing or video fluoroscopic swallowing study as the gold standard, the pooled sensitivity, specificity, positive and negative likelihood ratios, and diagnostic odds ratio were 0.89 (95% confidence interval [CI] 0.82–0.93), 0.59 (95% CI 0.39–0.77), 2.17 (95% CI 1.38–3.42), 0.19 (95% CI 0.13–0.29), and 11.49 (95% CI 5.86–22.53), respectively. When a cutoff of 3 was used, these values were 0.85 (95% CI 0.68–0.94), 0.82 (95% CI 0.65–0.92), 4.84 (95% CI 1.72–13.50), 0.18 (95% CI 0.07–0.46), and 26.24 (95% CI 5.06–135.95), respectively. Using EAT-10 cutoff values of 2 and 3, the areas under the curve were 0.873 (95% CI 0.82–0.93) and 0.903 (95% CI 0.88–0.93), respectively, showing good diagnostic performance. EAT-10 can be used as a preliminary screening tool for dysphagia. However, a cutoff of 3 is recommended for EAT-10 due to better diagnostic accuracy.