The Kazakh is one of the agglutinative language with more complicated morphological changes. Kazak stem and affix extraction have important significance for Kazakh information processing. In this paper, according to the morphological structure of Kazakh words, we applied a method to stem extraction, which is combined the lexical rules with statistical model. The stem extraction is carried out by using prefix dictionary, suffix dictionary, stem dictionary, statistical model dictionary and the rule base. Experimental results show that, in the statistical model, the method to extract the stem by using part of speech features is effective, in that, the word level accuracy and the stem level accuracy of this method reached 0.93% and 76.74% respectively.