A fast analytical method was introduced based on near-infrared (NIR) technology in this paper. The protein content was measured in short order using the near-infrared transmission spectroscopy (1000-1700nm) of milk. There were several waves of milk's NIR spectroscopy selected. By correlating the spectrum data of the waves selected and the protein content in milk, a calibration model was established. The protein content could be measured by importing the spectrum data to the calibration model. In this model there were several parameters, which were the spectrum data of the waves selected. Then, the method how to select the waves best was introduced and the characteristic waves of milk were selected by utilizing genetic algorithm. A partial least squares (PLS) regression model between the spectroscopy and the protein content was presented for milk samples, and the predictive repeatability was also researched.