Triptolide (TP) is the crucial active ingredient of Tripterygium glycoside tablets and has been shown to have a significant therapeutic effect on delayed-type hypersensitivity (DTH)-related diseases. However, due to its potential hepatotoxicity and nephrotoxicity, adverse reactions have often been observed in long-term treatment regimens. Therefore, it is meaningful to find metabolic markers for toxicity for early diagnosis. In this study, a feasible strategy using HPLC-HRMS method combined with multivariate statistical analysis to discover toxic potential markers of TP was developed. TP was used to treat a DTH mouse model at a therapeutic dose (45μg/kg) and toxic dose (900 μg/kg). The metabolic profiles of the liver, kidney and plasma were characterized by HPLC-Q/TOF MS. Significant differences in the metabolite profiles of the liver, kidney and plasma existed between the toxic and therapeutically dosed mice. Forty-six metabolites were identified and 27 of them may be related to toxicity based on a structure-toxicity prediction model. Using OPLS-DA analysis, the metabolite profiles between the two dose groups could be well distinguished. It was found that 18, 4 and 4 metabolic markers were altered in the liver, kidney and plasma, respectively; 15, 4 and 3 of these metabolic markers were predicted to be toxic. Two toxic markers detected both in mouse plasma and human liver microsomes following incubation with TP showed great potential as early diagnosis markers for TP hepatotoxicity and nephrotoxicity.