This study identified the metabolic biomarkers for different clinical phases of bipolar disorder (BD) through metabolomics. BD patients were divided into three groups: patients with BD and depressive episodes (BE, = 59), patients with BD and mania/hypomania episodes (BH, = 16), patients with BD and mixed episodes (BM, = 10), and healthy controls (HC, = 10). Serum from participants was collected for metabolomic sequencing, biomarkers from each group were screened separately by partial least squares analysis, and metabolic pathways connected to the biomarkers were identified. Compared with the controls, 3-D-hydroxyacetic acid and N-acetyl-glycoprotein showed significant differences in the BE, BH, and BM groups. This study suggests that different clinical types of BD share the same metabolic pathways, such as pyruvate, glycolysis/gluconeogenesis, and ketone body metabolisms. In particular, abnormal glycine, serine, and threonine metabolism was specific to BM; β-glucose, glycerol, lipids, lactate, and acetoacetate metabolites were specific to depressive episodes; the guanidine acetic acid metabolites specific to BH; and the acetic and ascorbic acids were metabolites specific to manic and BM. We screened potential biomarkers for different clinical phases of BD, which aids in BD typing and provides a theoretical basis for exploring the molecular mechanisms of BD.