Abstract: Mutual information based Chinese word segmentation and new terms extraction are typical statistics-based Chinese information processing technologies in recent 20 years．This paper discussed the mutual information distribution characteristics of frequent 2-gram,3-gram and 4-gram Chinese characters in a large corpus．The statistic results show two obvious findings as follows．First,there are no evident mutual information boundaries between Chinese word and phrase,which means it is impossible to distinguish Chinese words and phrases with either mutual information or frequency．Second,the mutual information of words,phrases and illegal Chinese strings are mixed together,which drama-tically affects the precision of statistics-based Chinese information processing technology．These two findings show that Chinese word extraction and segmentation only based on statistic technology still face great challenges.