Virtual screen by molecular docking has been successfully applied in lead discovery. However, it suffers from huge computational time, difficulties in final compounds selection, and correct binding pose prediction. We have developed a combined comprehensive “virtual screening” strategy and successfully applied to identify antagonist of HLA-DR4 from a pool of 4,000,000 chemical compounds based on the complex crystal structure of human HLA-DR4 bound with influenza ha antigen peptide. Five compounds were selected and tested in a cell based assay and two non-peptide compounds exhibited 17μM and 31μM inhibitory effects in the cell competitive test with the peptide collagen II against HLA-DR4, which function equivalently or better than those peptide antagonists. These compounds are among the first examples of potent, non-peptide antagonists for the HLA-DR4, which should be useful for further development of non-peptide anti-rheumatoid arthritis drugs.