Non-alcoholic fatty liver disease (NAFLD) is the most prevalent liver disease worldwide. Chronic activation of endoplasmic reticulum stress (ERS) in hepatocytes may promote the development of NAFLD, yet endoplasmic reticulum stress-related genes (ERSGs) have not been studied in NAFLD. Our aim is to study the relationship between ERSGs and the immune microenvironment of NAFLD patients and to construct predictive models. We screened 48 endoplasmic reticulum stress-related differentially expressed genes (ERSR-DEGs) using data from two GEO datasets and the GeneCards database. Enrichment analysis revealed that ERSR-DEGs are closely associated with immune-related pathways and functions. The immune infiltration profile of NAFLD was obtained by single sample gene set enrichment analysis (ssGSEA). There were significant differences in immune cell infiltration and immune function between NAFLD group and control group. Using 113 NAFLD samples, we explored two molecular clusters based on ERSR-DEGs. A five-gene SVM model was selected as the best machine learning model, and a nomogram based on five-gene SVM model showed good predictive efficiency. The mRNA expression levels of POR, PPP1R15A, FOS and FAS were significantly different between NAFLD mice and healthy mice. In conclusion, ERS is closely associated with the development of NAFLD. We established a promising and SVM-based predictive model to assess the risk of disease in patients with ERS subtypes and NAFLD.