Abstract: Wavelet analysis is utilized on eliminating noises and interp reting zero crossings of welding current, arc voltage and arc sound in CO₂ arc welding process, which are collected by experiment system. Extracted energy of arc sound in different frequency ranges is as characteristic vectors that can indicate changes in the welding processing, The correlation between characteristic class and welding sprays is studied by statistic test to evaluate and reduce dimensions of characteristic class. The base to fulfill on-line diagnose of welding process has been advanced in this paper.