Purpose. Esophageal cancer (EC) is a lethal digestive tumor worldwide with a dismal clinical outcome. Endoplasmic reticulum (ER) stress poses essential implications for a variety of tumor malignant behaviors. Here, we set up an ER stress-based risk classifier for assessing patient outcome and exploiting robust targets for medical decision-making of EC cases. Methods. 340 EC cases with transcriptome and survival data from two independent public datasets (TCGA and GEO) were recruited for this project. Cox regression analyses were employed to create a risk classifier based on ER stress-related genes (ERGs) which were strongly linked to EC cases’ outcomes. Then, we detected and confirmed the predictive ability of our proposed classifier via a host of statistical methods, including survival analysis and ROC method. In addition, immune-associated algorithm was implemented to analyze the immune activity of EC samples. Results. Four EGRs (BCAP31, HSPD1, PDHA1, and UBE2D1) were selected to build an EGR-related classifier (ERC). This classifier could distinguish the patients into different risky subgroups. The remarkable differences in patient outcome between the two groups were observed, and similar results were also confirmed in GEO cohort. In terms of the immune analysis, the ERC could forecast the infiltration level of immunocytes, such as Tregs and NK cells. Conclusion. We created a four-ERG risk classifier which displays the powerful capability of survival evaluation for EC cases.