The determination of the main components of milk is an important criterion for evaluating the quality of milk. Relevant national departments have formulated a series of relatively detailed specifications to ensure the quality and safety of milkand other dairy products. However,thetraditional detection methods are often complex, time-consuming and labor-intensive. Some even cause environmental pollution, making it difficult to meet the rapid detection needs of contemporary dairy production and consumption. In this study, the portable attenuated total reflectionFourier transform infrared spectroscopy (ATR-FTIR) technique was combined with relative humidity (RH) control system to establish a method to measure the infrared spectra of different kinds of milk under the condition of continuous decline of RH. This method provides a new way for non-destructive testing, classification and quality analysis of milk products. The main contents include:(1)Selecting five types of milk of YiLi brand (pure milk, Zhennong milk, skimmed pure milk, high-calcium low-fat milk, and Shuhua milk) as research objects. Whose infrared spectra in the process of evaporation and concentration were collected under continuous decline of RH, and the peak position attribution and qualitative analysis of main nutritional components were carried out. It only takes a few microliters of milk samples for us to obtain the spectral information of the main components, such as water, carbohydrates, fats, and proteins, during the sample concentration process in a short time and achievea relatively comprehensive characterization of the chemical components of milk; (2) Using NWUSA software to build the model and process the infrared spectral data, choosing 4 000～400 cm-1 band as the variables to perform PCA and evaluating the identification ability of the model for different types of milk, it shows that the data of PCA process are well aggregated in the same group. The coordinate axes in different groups are far apart, indicating that the model selection is both reliable and representative. A total of 75 milk samples were used in the experiment, in which the production date and place were random factors, and the type and brand of milk were fixed factors. The results show that the proposed method has the advantages of simple operation, sensitive response, high spectral quality and non-destructive measurement, which is suitable for in-situ, rapid and non-destructive identification and analysis of milk and other dairy products.