Pronunciation error feedback plays an important role in the CALL (Computer Assisted Language Learning) system. Accurate error feedback with detailed and appropriate guidance can make it easier for learners to correct their mistakes and to avoid similar mistakes in the future. In order to provide visualized pronunciation error feedback automatically and flexibly, in this paper, we proposed a three-dimensional tongue modeling method based on PCA. To do so, firstly, pronunciation data, which covers all 28 Japanese phonemes, was constructed from a three-dimensional physiological articulatory model. Then, the critical points of the three-dimensional tongue surface were picked and added to a sample matrix. Finally, the principal components were derived by PCA (Principal Component Analysis) method, and the tongue could then be reconstructed according to the analysis result. Based on the above work, we constructed a prototype of the feedback presentation system by combining the tongue with the palate and jaw.