In this study, the process to synthesize a human face based on the information of words is defined as a mapping from a word space, which is composed of the words expressing dimensions and shape of facial elements, into a physical model space where physical shape of the facial elements are formed. By introducing a concept of mapping, the use of whole words existing in the word space makes it possible to synthesize a human face based on free and uninhibited description. Furthermore, we have only to make 3-dimensinal physical models corresponding to the words that are selected as training data to identify a mapping function. The others are made through the mapping. Finally, we inspect the validity of the mapping function that is obtained in this study.