Support Vector Machine (SVM) is a kind of pattern classification method. Because learning of SVM is based on quadratic optimization, SVM can easily handle large scale problems with many training examples. But learning of SVM takes a long time for such large problems. In this paper, we parallelize Support Vector Machine using divide and conquer method. In addition, we discuss accuracy of parallel SVM, and evaluated learning time. Experimental results show that parallel SVM achieves faster learning with keeping accuracy than that of sequential SVM.