Psychiatric disorders, such as major depression disorder (MDD), bipolar disorder (BP), and schizophrenia (SZ) are severe mental illnesses associated with morbidity and life-long disability for sufferers. Abnormal behavior and disturbed cognition, often assumed to represent psychiatric disorder, may actually result from some form of abnormal brain disease that can be detected by means of one or more biomarkers. However, heterogeneity of psychiatric disorder illness course complicates clinical decision-making. In recent years, the search for psychiatry-relevant biomarkers of major depression, bipolar disease and schizophrenia has intensified. In this study, quantitative targeted proteomics was performed on psychiatric disorder patients using liquid chromatography-mass spectrometry. We used plasma samples (40 normal control, 50 MDD, 50 BP, 50 SZ) for construction of biomarker panels for differential diagnosis of MDD, BP and SZ. In additions, we compared proteins expression levels of psychiatric disorder patients with that of normal control. For multi-marker panel, AUCs based on the machine learning algorithm in BP vs. SZ showed the highest value with 0.990, 0.882 in training and test set, respectively. These results confirm that clinically useful differential diagnosis in MDD, BP, and SZ, also improve on the conventional methods in developing such models.