The online non-contact in-situ spectral detection technology of milk powder composition is very beneficial to dairy production enterprises, which is very important for processors to evaluate the quality of products in real-time. Raman spectroscopy is a very commercial potential in-situ spectral detection technology. This paper analyzes the feasibility of in-situ Raman spectroscopy in detecting the main components of milk powder (fat, protein, lactose, etc.). The Raman spectrum experiments of skimmed milk powder and whole milk powder with different mixing ratios were carried out, and the statistical laws of Raman spectrum characteristic peaks and fat content were studied. The prediction model based on the partial least squares regression method was established, and the fat content was predicted based on the regression model. The determination coefficient of the regression model was more significant than 0.9886, and the root mean square error was less than 0.6248. In conclusion, in situ, Raman spectroscopy has potential commercial application value for the online evaluation of milk powder quality.