Abstract： Objective To innovate an early, rapid and efficient approach to the pmgnestic evaluation of sep-sis in order to lower the mortality. Method Forty-five septic rats, induced by cecal ligation and puncture, were divided into surviving group (n=23) and non-survival group (n=22) on six days after onset of sepsis. Serum samples were taken from septic and sham-operated rats (n=25) at 12 hours after surgery. HPLC/MS assays were performed to acquire the serum metabolic profiles, and radial basis function neural network (RBFNN) was em-ployed to build predictive model for prognostic evaluation of sepsis. Results The principal component analysis al-lows differentiating the rots of survive,non-survive and sham-operated from one another in respect of the pathologic characteristics. Six metabolites, linolenic acid, linoleic acid, oleic acid, stearic acid, docosahexaenoic acid and do-cosapentaenoic acid, related to the outcomes of septic rats were then structurally identified. A RBFNN model for outcome predication was built based upon the metabolic profile data from rat sera with the sensitivity of (96.1 ±3.6)% (n=10) and specificity of (91.0±4.3)% (n=10). Condusions HPLC/MS-based metabonomic approach combined with pattern recognition permits accurate outcome prediction of septic rats in the early stage. The proposed approach has advantages of rapid, low-cost and efficiency, and is isph-ing to be applied in clinical prognostic evaluation of septic patients.