Combining modified ensemble empirical mode decomposition (MEEMD) and chaotic oscillator, a distribution network fault weak signal detection method was proposed. At first, a neural network prediction model was set up, which was used to conduct the short term prediction of the zero sequence currents of the lines in the distribution network and filter out the background signal in the fault signal. Then, in order to detect the weak 5th harmonic fault signal when the single phase grounding fault happens in the distribution network, the MEEMD method combining multiscale permutation entropy and complete ensemble empirical mode decomposition (CEEMD) was proposed to process the fault signal in which the background signal was filtered out. The first intrinsic mode function was extracted and used as the input of the chaotic oscillator. The chaotic oscillator has high sensitivity to the external driving force signal with the same frequency as that of the internal driving force signal. The distribution network fault line selection was completed with the phase diagram outputted by the chaotic oscillator.