Taking the whole blood obtained from daily clinical inspection as experimental materials, triglyceride concentration of serum is determined by traditional clinical method: glycerol phosphate oxidase-peroxidase-4-amino antipyrine-phenol method. Near infrared spectroscopy of all serum samples are collected at the same time, with the range between 12500~4000 cm-1. A quantitative model of serum triglyceride concentration by near infrared spectroscopy is established by combining with partial least squares (PLS) method. Experimental results show that the proper variable range is 10796.2~8246.6 cm-1, and with the multiplicative scatter correction as spectrum pretreatment method, the quantitative PLS calibration model for serum triglyceride is established. The correlation coefficient R2=0.9454, cross-validation calibration standard deviation (RMSECV) is 0.146 for the calibration set; the predicted standard deviation (RMSEP) is 0.151, and correlation coefficient R2=0.9068 for the test set. The results show that, by the application of near infrared spectroscopy combined with partial least squares method, the serum triglyceride quantitative model is successfully established with high accuracy and stability, which can be used for non-destructive testing of triglyceride content in the unknown serum samples.