Paraquat (PQ) has caused countless deaths throughout the world. There remains no effective treatment for PQ poisoning. Here we study the blood metabolome of PQ-poisoned patients using ultraperformance liquid chromatography coupled to quadrupole time-of-flight mass spectrometry (UPLC/Q-TOF MS). Patients were divided into groups according to blood PQ concentration. Healthy subjects served as controls. Metabolic features were statistically analyzed using multivariate pattern-recognition techniques to identify the most important metabolites. Selected metabolites were further compared with a series of clinical indexes to assess the prognostic value. PQ-poisoned patients showed substantial differences compared with healthy subjects. Based on variable of importance in the project (VIP) values and statistical analysis, 17 metabolites were selected and identified. These metabolites well-classified low PQ-poisoned patients, high PQ-poisoned patients, and healthy subjects, which was better than that of a complete blood count (CBC). Among the 17 metabolites, 20:3/18:1-PC (PC), LPA (0:0/16:0) (LPA), 19-oxo-deoxycorticosterone (19-oxo-DOC), and eicosapentaenoic acid (EPA) had prognostic value. In particular, EPA was the most sensitive one. Besides, the levels of EPA was correlated with LPA and 19-oxo-DOC. If EPA was excessively consumed, then prognosis was poor. In conclusion, the serum metabolome is substantially perturbed by PQ poisoning. EPA is the most important biomarker in early PQ poisoning.