The problems arising in the modeling and coding of digital Chinese character patterns for noiseless compression purposes are discussed. The modeling is intended to capture the maximum redundancies of the source under the consideration of relevant parameters, In this study, a redundancy-gathering algorithm is proposed in which a 2-D parsing tree is used, and the nodes to extend the tree are ranked according to the maximum redundancy gathered. Hence, the modeling represented by the tree can achieve great performance in the small number of nodes. The algorithm is then applied to evaluate the performance of arithmetic coding for digital Chinese character patterns. The results are compared to those of the traditional compression methods. In the same complexity, we find that our algorithm can improve the coding efficiency as high as 25.54%, 30.23%, 32.17%, 15.90%, and 27.58%, for five Chinese fonts, respectively.