Abstract: Based on the support vector machine (SVM) 1-v-1 and DDAG (decision directed acyclic graph) multi-classified algorithm, a new approach to the soft fault diagnosis of analog circuits is presented. The DDAG is a newly developed learning system on 1-v-1 basis in which the idea of directed acyclic graph of the graph theory is introduced to combine SVM subclassifiers together. Then, the SVM classification results by using different kernel functions are compared experimentally with each other. The simulation results show that the diagnosis accuracy is up to 99% if using the DDAG support vector machine algorithm. In this way the higher diagnostic accuracy is available.