Abstract: The detection of distribution transformer partial discharge by using ultrasonic wave has the advantage of high detection sensitivity, good anti-interference performance. In order to study the discharge characteristics of different types of flaws in distribution transformer partial discharge with ultrasonic testing, we set up an ultrasonic testing platform and constructed four typical oil-paper insulation defect model and studied their time-domain waveform characteristics. The results shows that combining the FIR filter and wavelet de-noising could remove the noise signal in detection. The results of time-domain detection show that different type of defects has different time-domain waveform feature, with which we could distinguish the type of flaws preliminarily.