Psoriasis is a chronic inflammatory skin disease that have bad effects on the quality of life of the patients. As psoriasis is intractable and its cause is difficult to discover, Traditional Chinese Medicine is proved in China to be a more effective medical way. In Chinese Medicine, decision on prescription is based on ZHENG rather than disease. Only after successful differentiation of ZHENG, can effective treatment of TCM be possible. As many papers in ZHENG classification modelling were reviewed, one common characteristic was found that although the original medical records were written and stored in text format, most experiments in these papers use data in a structured format which was extracted from its original text format. Therefore, whether or not full usage of information is extracted from original text should be considered seriously in building ZHENG classification. In this paper, machine learning methods were used to evaluate four feature extraction methods' capability in extracting useful information for psoriasis ZHENG classification from medical texts. The experiment result revealed that feature extraction has great influence on ZHENG classification and doctors' segmentation of medical case text, as the punctuations indicate, contains some information that dictionary does not contain and but is essential in ZHENG identification, such as the group of symptoms, degree words and so on. What's more, models with features from bow perform better than that from word2vec, which may illustrate that the sequence of words in medical texts has little impact on the classification of ZHENG.