Existing quick or early detection and identification methods for short fault are based on the sharp increase of the short current to extract the fault features in low-voltage system. They all have a technical bottleneck problem of miscarriage of justice, which is affected by the noise of the distribution system and the non-fault running state interference caused by different load starting or switching. On the basis of establishing the simulation model of the single-phase short in the low-voltage system, this paper analyzes waveforms of the faulty branches terminal voltage in full angle range and concludes the characteristic which faulty branches terminal voltage almost drops to zero at the moment of short circuit and remains near zero zone before fault elimination. And based on the above analysis, this paper introduces the wavelet packet decomposition method and combines wavelet packet decomposition method with instantaneous value of terminal voltage so as to eliminate the interference of noise and different types of load startup etc. Simulation and experimental not only verify the effectiveness and fast of the terminal voltage early detection method for short circuit but also solve the problem of the reliability of the existing methods for early detection of short circuit current.