Most two-components Chinese characters are pictophonetic characters, which have one component to indicate the sound, named as the sound-element, while the other to convey the meaning, named as the meaning-element. With the development of speech sounds and character radicals, sound-elements cannot indicate the pronunciation precisely as before. In this paper, we will use similarity matrices to calculate the pronunciation indication of sound-elements, and classify pictophonetic characters with cluster analysis method. It is innovative to apply mathematical model to the research of Chinese characters, and our research will not only provide reference to normalization of modern Chinese characters, but also help to achieve better results in teaching or learning Chinese as a second language.