Abstract: To explore the proper sensing techniques which should be able to tell the size of the keyhole in real time during the plasma welding processes, the signals of arc voltage and current under the conditions of different keyhole sizes were sampled into the computer, and processed to find characteristics related to keyhole size using ‘FFT’ and ‘STFT’ techniques on the signals. After analyzing of the data gained the features of the local maximum of the spectrum peeks were correspondent to the keyhole size when the frequent region of spectrum was form 2.5 kHz to 3.5 kHz. It was found that the fluctuation of the amplitude of the local maximum of the spectrum peeks slows down as the keyhole expands and the relevant frequency gradually shifts. Proper arithmetic have been bevised to quantify the relationship between the amplitude ‘pf’ and the frequency ‘pff’ of the local maximum of thespectrum peeks and the keyhole size. Then welding experiments were carried out using the work pieces of thickness changing to test the validity of the detection of keyhole size by ‘pf’ and ‘pff’ signals. The physical relationship between the arc signals of ‘pf’ and ‘pff’ and the ‘keyholeeffect’ has been explained preliminarily. At last,‘pff’ signal has been selected to act as the sensing signal of the keyhole size, and this technique can provide the necessary feedback information for the automatic control system on the precise plasma welding of work pieces of great and modest thickness.