Automatic text classification of web texts in Asian languages is a challenging task. For text classification of Thai web pages, it is necessary to cope with a problem called word segmentation since the language has no explicit word boundary delimiter. While a set of terms for any texts can be constructed with a suitable word segmentation algorithm, Thai medicinal texts usually has some special properties, such as plentiful of unique English terms, transliterates, compound terms and typo errors, due to their technical aspect. This paper presents an evaluation of classifying Thai medicinal web documents under three factors; classification algorithm, word segmentation algorithm and term modeling. The experimental results are analyzed and compared by means of standard statistical methods. As a conclusion, all factors significantly affect classification performance especially classification algorithm. The TFIDF with term distributions, as well as SVM, achieves high performance on non-segmented and segmented Thai medicinal web collection as they efficiently utilize technical terms.