The technology of tongue near-infrared reflectance spectra was used for human serum total protein (TP) content of noninvasive testing for the first time. Reflectance spectrum on the tongue tips of 58 volunteers was collected, and the biochemical values of serum total protein were recorded at the same time. The samples were separated into two parts: training set and prediction set. Two prediction models were established using PCA combined with BP neural network and PLS. Using PCA-BP model to predict the prediction set, the average relative error is 7.35%, RMSEP was 6.3771 g x L(-1), and the correlation coefficient was 0.9021. Using PLS model to predict the prediction set, the average relative error is 4.77%, RMSEP was 0.1304 g x L(-1), and the correlation coefficient was 0.9718. It was approved that reflectance spectra of tongue can be used to predict TP accurately and noninvasively.