The limitation of exiting instruction-words software birthmark was high complexity, this paper introduced the idea of characteristics selection in text classification into the field of software birthmark, and proposed an instruction-words software birthmark selection method based on CHI ( statistics). The birthmark selection algorithm build sample program collections for protected program and extracted instruction-words from sample programs according to the instruction-words library firstly. After that, it calculated the statistics for each instructions-word in order to measure the correlation between instruction-words and program. Experiments showed that the birthmark selection algorithm could effectively bring down the scale of data, and significantly improved the credibility and robustness of the birthmark.