The quality of agricultural products and food has always been one of the focuses of attention. The quality and safety of agricultural products and food are related to people’s health and related to social stability and even national security. In recent years, the safety incidents caused by the unqualified quality of agricultural products and food have attracted all social circles’ attention. The supervision of the quality of agricultural products and food has been the key point even difficulty in analysis and detection for a long time. Given a large population, the consumption of agricultural products and food is enormous in China. In the face of such a large number of the needs of non-destructive and rapid detection of agricultural products and food quality, spectroscopy analysis can provide a good solution for the non-destructive and rapid detection for agricultural products and food with the characteristics of fast, non-destructive, efficient, environmentally friendly, on-site testing. However, due to the large amount of data used in the traditional spectral analysis, it is time-consuming in developing calibration models and difficult to complete the online, high-throughput, non-destructive and rapid detection of the large number of agricultural products and food quality. On the other hand, the calculation of such a large number of data has also become one of the main bottlenecks limiting the efficiency of spectral analysis instruments, and the calculation of a large number of data also puts forward very high requirements for the hardware configuration of the instruments, which will increase the application cost of spectral analysis technology indirectly. In recent years, key variable selection has emerged and become a new hotspot of spectral analysis. According to the selection, calibration models can be developed by a few numbers of the key variables, which are of almost the same accuracy to the models developed by the full spectra. Thus it can improve the analytical instruments’ working efficiency effectively and reduce the application cost of the spectral analysis technology. It will also provide reliable technical support for the high-throughput detection of agricultural products and food quality and provide the scientific and technological support for meeting the increasing demand of the people for a better life. In this paper, the applications of spectral key variable selection in the non-destructive detection of grain and grain crops, vegetables, fruits, cash crops, meat, food quality and safety were reviewed. With summarizing others’ works in recent years, the applications of spectral key variable selection technology were summarized from the aspects of selection method, application scope, application effect, and so forth. Finally, the application of spectral key variable selection technology in non-destructive detection of agricultural products and food quality prospected from the aspects of the characteristics and trends of the variable selection methods, the stability and reliability and the practical significance of the selected variables.