OBJECTIVES: Endocytoscopy (EC) is an ultra-high magnification endoscopy designed to provide in vivo histologic assessment. This study aimed to evaluate the diagnostic yield of the newly developed fourth-generation EC for esophageal squamous lesions by using a modified EC classification. METHODS: A total of 2548 EC images of 57 esophageal targeted areas between June 2015 and October 2017 were retrospectively collected. Two lesions with low-quality images were excluded. Only EC images were independently reviewed by two expert and two non-expert endoscopists. The lesions were classified according to a three-tier modified EC classification. We used a multilevel logistic regression to analyze the data. RESULTS: The sensitivity and specificity of diagnosing non-squamous cell cancer (SCC) vs SCC were 82.5% and 83.0% by the experts; 90.1% and 75.0% by non-experts. The interobserver agreement among the four raters was good (kappa statistic 0.59). The diagnostic accuracy of experts and non-experts was similar (P = 0.16 for specificity and P = 0.20 for sensitivity). The sensitivity and specificity of EC for non-neoplasia vs neoplasia were 88.7% and 74.6% by experts; 90.3 and 52.1% by non-experts. The interobserver agreement among the four raters was moderate (kappa statistic 0.44). The specificity of experts was higher compared to non-experts, although the difference did not reach statistical significance (P = 0.08 for specificity and P = 0.93 for sensitivity). CONCLUSIONS: Fourth-generation EC offers acceptable diagnostic accuracy and reliability in both experts and non-experts, especially when diagnosing SCC lesions.