A small non-invasive measurement system for human blood glucose has been developed, which can achieve fast, real-time and non-invasive measurement of human blood glucose. The device is mainly composed of four parts, i.e. fixture, light system, data acquisition and processing systems, and spectrometer. A new scheme of light source driving was proposed, which can meet the requirements of light source under a variety of conditions of spectral acquisition. An integrated fixture design was proposed, which not only simplifies the optical structure of the system, but also improves the reproducibility of measurement conditions. The micro control system mainly achieves control function, dealing with data, data storage and so on. As the most important component, microprocessor DSP TMS320F2812 has many advantages, such as low power, high processing speed, high computing ability and so on. Wavelet denoising is used to pretreat the spectral data, which can decrease the loss of incident light and improve the signal-to-noise ratio. Kernel partial least squares method was adopted to build the mathematical model, which can improve the precision of the system. In the calibration experiment of the system, the standard values were measured by OneTouch. The correlation coefficient between standard blood glucose values and truth values is 0.95. The root mean square error of measurement is 0.6 mmol·L-1. The system has good reproducibility.