For a quick，objective and non-invasive examination of the tongue for pathological information,a near infrared spectral classification examination model,based on probability neural network tongue diagnosis after normalized reflectivity pretreatment,is proposed.With the collection of spectral data from the tip of the tongues from healthy people,fatty liver patients,and hepatitis patients,32 sample data from each group are collected after normalization reflectivity pretreatment.24 sample data from each case are selected for the construction of the models.Then the remaining 24 sample data are adopted for prediction.Also the influence of the spread of radial primary function upon the internet is analyzed.The conclusion is that the predictability and generalization of the network models work well when the spread ranges from 0.0065 to 0.0077.The predictions,based on the model,toward the samples are 95.8% correct.Experimental results show that the near infrared spectral classification of the tongue diagnoses in accordance with probabilistic neural network is feasible,and it is proved that spectral information of the tongue surface can objectively reflect human pathological information,so the preliminary screening which is rapid and noninvasive is probable with the help of this method.