Text classification is one of the most actual among the natural language processing problems. In this paper the application of word-based PPM (Prediction by Partial Matching) model for automatic content-based text classification is described. Our main idea is that words and especially word combinations are more relevant features for many text classification tasks. Key-words for a document in most cases are not just single words but combination of two or three words. The main result of the implemented experiments proved applicability of word-based PPM models for content-based text classification. Although in some cases the entropy difference which influenced the choice was rather small (several hundredths), most of the documents (up to 97%) were classified correctly.