A miniature wearable Raman spectroscopy system used to achieve noninvasive detection of human blood glucose level was developed. A thallium-doped grin lens was employed as the collection lens and a specially designed wearable fiber optic probe was employed to help the stable and convenient collection of Raman spectrum. The glucose solution, 11 rats and 10 healthy human were studied as subjects. In addition, a method of quantitative analysis of Raman spectrum was proposed which using the peak area as the main reference factor while the peak intensity as the auxiliary reference factor to calculate the target concentration. A non-linearized multivariate dominant factor-based partial least squares model was built for different samples to predict glucose level. The results show that the accuracy are 98.1%, 89.3% and 84.4% for glucose solution, rats and human subjects. The system has the advantages of more compact structure, lower cost, better testing stability and convenient for human body to wear, and is feasible and repeatable to achieve the noninvasive detection of human blood glucose accurately.