In order to diagnose fatty liver noninvasively, rapidly and accurately, this article presented a new method based on spectroscopy to diagnose fatty liver. This method is non-invasive, rapid. Because tongue can objectively reflect physiological and pathological changes, so this experiment collected reflectance spectrum on the tongue tips of volunteers at first, then analyzed the above data, calculated the normalized reflection ratio, and built a three-layer BP network model. Thirty two healthy people and 44 fatty liver sufferers were chosen randomly from the total 115 samples and their data were input into neural network, then the data of unknown samples of 14 healthy and 25 fatty liver ones were input into the model, and the classification accuracy was 89.7%. This result approved the feasibility of using spectroscopy to diagnose fatty liver. Meanwhile, the result showed that spectral method can reflect the information of organization and tiny circulation taken by tongue more objectively. This method may provide a fast and simple diagnostic tools for clinic, and also can provide a reference to the syndrome measurement of the traditional Chinese medicine.