For a quick and noninvasive examination of the tongue for potential hepatitis patients, a study was conducted on the relation between the reflectivity of near infrared spectrum on the tongues of the healthy people and the hepatitis patients. Spectral data, 25 items for each case, are to be collected from the left and right side tongue, left and right sublingual venae, and tip of the tongue from the healthy people and the hepatitis patients. Then a three-layer neural network structure was established with all the data input after normalization reflectivity pretreatment. With the establishment of a BP neural network model, 40 data from each part of the body were selected as training samples. The rest 10 were adopted for prediction, which later was proved to be 100% correct with relative deviation values of less than 0.2. The research findings show that the proposed application of BP neural network in the spectral noninvasive examination of the tongue for identification of different case diagnosis is important for reference.