ABSTRACT: Thyroid eye disease (TED) is an orbital autoimmune disease which severely affects vision, appearance, and quality of life. TED is the most common extrathyroidal manifestation of Graves' disease, which is mainly featured by hyperthyroidism. Current assessments of TED severity and activity, do not effectively guide regimen, especially the selection of target agents. Peripheral blood samples from 161 TED patients and 20 Graves' hyperthyroidism (GH) patients without eye disease were collected for transcriptome sequencing, and corresponding clinical information was systematically compiled. We then comparatively derived a unique gene expression profile for TED compared to GH, and based on this, unsupervised clustering of TED patients using ConsensusClusterPlus was conducted. Next, immune infiltration analysis, weighted gene co-expression network analysis (WGCNA), hub gene screening, and drug sensitivity analysis were performed for different clusters. TED exhibited elevated of autoantigens, immune responses, cellular proliferation, and metabolic disorders than GH. A novel two-cluster classification scheme was established for TED, independent of classical assessment of severity and activity. Cluster 1 patients received more intravenous glucocorticoids and surgeries but experienced less pain and excessive tearing. They had higher peripheral blood lymphocytes, including CD4+ T cells, CD8+ T cells, T follicular helper cells, natural killer cells, and B cells, than cluster 2 patients. WGCNA identified cluster-related gene modules, and least absolute shrinkage and selection operator (LASSO) regression analysis yielded six hub genes (DDR2, LRP1B, NTRK2, PALB2, PTPRD, and SLC25A6) for cluster 1 and three hub genes (MON2, SNORA74B, and STX16) for cluster 2 as potential biomarkers. Considering current TED therapies, cluster 1 patients may prefer insulin-like growth factor 1 receptor (IGF-1R) antibodies, while cluster 2 patients may favor glucocorticoids. We described the specific gene profile of TED compared to GH and built a new molecular classification of TED patients into two clusters with different immune infiltration, hub genes, and drug sensitivity. It paves the way for improved application of precision medicine in TED.