The principles and methods were introduced for measuring soil organic matter content with the aid of spectrum data diffusely reflected by the soil. The arrangement scheme, light system, hardware and software were carefully designed. In order to decrease the measurement errors, two other signals from different wavelengths of light were used for including the variation of the soil H/F value (the ratio of humic acid to fulvic acid), not as the pioneers only used one wavelength of light; beside sieving the soil samples, a comprehensive procedure was used——soil types were discriminated first and then the predicting model was built with the aid of partial least-squares regression method which was robust to the multi-collinearity among the variables . As a result, a better light system and a predicting model between soil organic matter and the reflected spectrum data were developed. The result showed that compared the errors of the measurements gained by this system with those by the standard methods, the precise of this system is within the permissible range in real operation.