When　a　high　impedance　fault　(HIF)　occurs　in　a　distribution　network,　the　detection　efficiency　of　traditional　protection　devices　is　strongly　limited　by　the　weak　fault　information.　In　this　study,　a　method　based　on　S-transform　(ST)　and　average　singular　entropy　(ASE)　is　proposed　to　identify　HIFs.　First,　a　wavelet　packet　transform　(WPT)　was　applied　to　extract　the　feature　frequency　band.　Thereafter,　the　ST　was　investigated　in　each　half　cycle.　Afterwards,　the　obtained　time-frequency　matrix　was　denoised　by　singular　value　decomposition　(SVD),　followed　by　the　calculation　of　the　ASE　index.　Finally,　an　appropriate　threshold　was　selected　to　detect　the　HIFs.　The　advantages　of　this　method　are　the　ability　of　fine　band　division,　adaptive　time-frequency　transformation,　and　quantitative　expression　of　signal　complexity.　The　performance　of　the　proposed　method　was　verified　by　simulated　and　field　data,　and　further　analysis　revealed　that　it　could　still　achieve　good　results　under　different　conditions.