nowadays improvement of hardware and software tools redounds to ease of vast data storing. Day by day the number of text documents is increasing, e-mail, web pages, texts, news and articles are only part of this range of increasing. Hence, the need for text mining techniques like automated methods for text justify; LINE-HEIGHT: normal; MARGIN: 0in 0in 0pt" and its related stem are stored in some kind of structured form. Consequently, for each stored word, we find its stem. However, the approach needs more space. Also, for each new word, table must be updated manually. In statistical methods, through a process of inference and based on a corpus, rules are formulated regarding word formation. This approach does not require any linguistic knowledge whatsoever, being totally independent of the morphological structure of the target language. Stemming is used for improvement of performance of text mining techniques. It is also used for space dimension reduction, because the feature space includes tens of thousands of words that will cause the next processes of system be impossible. Different methods to stemming have been designed in various languages each with advantages and disadvantages. In Persian also some algorithms for stemming have been proposed that have their own advantages and disadvantages, but there is not a general method for text mining in Persian that has high performance. For improvement of stemming in Persian in this thesis, two new methods are presented. The first method is based on study of Persian morphological structure. The proposed approach is a hybrid method. In this method a lookup table and automata are used for finding stems. This method is a static method and it lacks flexibility. So it has some errors in stemming. Second approach also like first stemming method, is a hybrid method. First step of this method uses a lookup table. Second step of this method implemented with decision tree algorithm. Since learner methods are dynamic, some parts of disadvantages will cover. In the end, for comparison of performance, one of general Persian stemmers has been chosen. In this thesis also a complete preprocessing method for Persian documents is proposed. For examination of performance, text justify; LINE-HEIGHT: normal; MARGIN: 24pt 0in 0pt" dir=ltr Keywords: Text Mining, Text 0in 0in 10pt" dir=ltr