Serum tumor markers for the diagnosis of esophageal squamous cell carcinoma (ESCC) have low sensitivity. This study aims to identify new serum markers for ESCC diagnosis from RNA sequencing (RNA-seq) data. RNA-seq was performed using six pairs of ESCC and matched normal tissues. The candidates for ESCC were screened from the differentially expressed genes. The candidates were analyzed by ELISA from the serum of a test group and a validation group. Real-time PCR, Western blotting and immunohistochemistry were used to detect the expression of the candidates in tumor cell lines and tumor tissues. Ten genes were selected from the RNA-seq data. Serum levels of ADAM12, CHI3L1, MMP13 and SPPl were significantly higher in the ESCC patients than in the healthy controls. A diagnostic model combining CHI3L1, MMP13, and SPP1 was established. The area under the curve (AUC) values for serum CHI3L1, MMP13, and SPP1 and the diagnostic model for discriminating ESCC patients from controls were 0.732, 0.881, 0.661 and 0.928, respectively. In the validation cohort, the AUC values were 0.753, 0.789, 0.696 and 0.843, respectively. Moreover, the AUC of the model for classifying patients with early ESCC was 0.918 in the test group and 0.857 in the validation group. Overexpression of CHI3L1, MMP13 and SPP1 was observed in the tumor cell lines and tissues. The diagnostic model composed of CHI3L1, MMP13 and SPP1 discriminates ESCC patients with high sensitivity. Our data highlight the potential of this diagnostic model for the noninvasive diagnosis of ESCC.