Abstract: A new method was studied for determining the classification of dairy products based on spectrum analysis and Hidden Markov Model ( HMM) .Firstly, the spectrum data were collected, which sampled from 4 kinds of dairy product.Secondly, wavelet transform method, multi-point smoothing method and multivariate scattering correction method were used to preprocess spectral data, and the main characteristics of sample data were extracted by principal component analysis ( PCA) . Then, the processed data was divided into two collections, part of which was used to train the Hidden Markov classification model ( HMM) and the residual data was tested.The experiment results under 15 processing conditions showed that different pretreatment methods and main feature dimensions of PCA could affect the detection accuracy of the classification model.The experimental average result was more than 99%. In conclusion, HMM could be used in dairy products classification and had a stable classification accuracy.