Utilizing Near-infrared Spectroscopy for non-invasive glucose concentration sensing has been a focusing topic in biomedical optics applications. In this paper study on measuring conditions of spectroscopy on human body is carried out and a series of experiments on glucose concentration sensing are conducted. First, Monte Carlo method is applied to simulate and calculate photons’ penetration depth within skin tissues at 1600 nm. The simulation results indicate that applying our designed optical probe, the detected photons can penetrate epidermis of the palm and meet the glucose sensing requirements within the dermis. Second, we analyze the influence of the measured position variations and the contact pressure between the optical fiber probe and the measured position on the measured spectrum during spectroscopic measurement of a human body. And, a measurement conditions reproduction system is introduced to enhance the measurement repeatability. Furthermore, through a series of transmittance experiments on glucose aqueous solutions sensing from simple to complex we found that though some absorption variation information of glucose can be obtained from measurements using NIR spectroscopy, while under the same measuring conditions and with the same modeling method, choices toward measured components reduce when complication degree of components increases, and this causes a decreased prediction accuracy. Finally, OGTT experiments were performed, and a PLS (Partial Least Square) mathematical model for a single experiment was built. We can easily get a prediction expressed as RMSEP (Root Mean Square Error of Prediction) with a value of 0.5-0.8mmol/dl. But the model’s extended application and reliability need more investigation.