To facilitate noninvasive detection of human triglyceride (TG) in blood, near infrared spectroscopy (NIRS) was applied to quantitatively analyze the TG level of single person. By optimizing the proper NIRS detection band (5 700～5 600 and 4 600～4 400 cm-1) in in-vitro experiment, preferably designing the fiber probe, and comparatively evaluating several preprocessing algorithms, we intend to further promote the non-invasive detection accuracy and stability of single-person TG level. Then we noninvasively collected 54 samples of spectral data from the same volunteer and made quantitative analysis for TG level. Savitzky-Golay (SG) combined with partial least squares (PLS) was confirmed to be the most robustness calibration model. The optimal analysis of the predictive 1 and 2 sets is as follows. The standard deviation (RMSEP) was 12 and 12.8 mg·dL-1, respectively; the relative standard deviation (RSD) was 16.25% and 17.33%, respectively. The prediction accuracy was ideal and able to be used for routine monitoring of single TG. In view of the well performance of SG-PLS model in non-invasive detection of single-person TG level and the trend of daily variation, NIRS analysis technology has potential for human TG non-invasive detection and daily management.