Abstract: Because of text's properties, it's so difficult to find perfect algorithms of embedding watermark into text documents, when compared with other media. Nowadays, text watermarking technologies is far behind other multimedia watermarking technologies such as images, audios, etc. A novel text digital watermarking technology based on graph theory is presented in this paper. Through changing its topology structures of a character or string, different figures are schemed out, which represent the same semanteme. Then those figures are mapped on graphs of graph theory, and the graphs or their properties should be correctly encoded in order to denote different watermarking information. The mathematic models of embedding and detecting watermarking are also described, and the experimental methods of robustness and visual influence, as well as the relative results are given. Finally, general rules of removing attack on text watermarking system is proposed, and the attack-resisting capability and methods of this text digital watermarking is analyzed in detail. According to the experimentation and analysis, the text watermarking technology has the following advantages: great capacity of watermarking, strong robustness, little visual influence and quite good attack-resisting capability. With this technology, it is more convenient to embed watermark into characters for ideograph languages such as Chinese and Korean, while it is suitable for embedding watermark into strings for alphabetic languages such as English and French.