The arc fault characteristics in low voltage distribution lines are experimentally researched in this paper. The Mallat algorithm is appointed to transform arc fault signal,the obtained detail components in different scales are compared to establish characteristic quantity. The results show that arc fault detail components drastically ascend compared to normal current,in addition the high scales detail components play part in noise elimination. Eventually the comparison between arc fault current and start current in detail components aimed at avoiding misjudgment shows that Mallat algorithm is effectively used to detect arc fault.