The invention discloses a detection method for a fault arc, which can improve the accuracy of distinguishing and judging the fault arc effectively, whether periodicity lose, asymmetry of positive and negative semicircles, an excessive flat shoulder part and an excessive changing rate of a current waveshape exist or not is analyzed by collecting current data of each period based on universal characteristics of the arc, so that whether an arc fault happens or not is judged. Compared with the prior art, the accuracy and the precision of judging the arc fault can be improved, and the unwanted operation rate of an AFCI (Arc-Fault Circuit Interrupter) product is reduced.