It's a hotspot to expend the research on support vector machine from a two-class issue to a multi-class one. Among all kinds of methods, Bintree multi-class text categorization algorithm based on support vector machine is more effective in training and sorting then others, and it works out the impartibility problem. So it is a good method. The dissertation systematically researches and analyses Bintree multi-class text categorization algorithm based on support vector machine, and improves it. That is, assemble first, and then sort them when the size of testing texts is too large. The aim is that after improvement the judgment of the testing text does not have to begin from the base crunode of Bintree, instead the testing text can be put into category function to be computed. The improvement can enhance the efficiency of text categorization and the probability of accurate categorization when the size of testing texts is too big and the quantity of sorted functions is too large.