Detecting　short-circuit　faults　is　a　critical　part　of　fault　processing　in　distribution　networks,　which　is　useful　for　the　subsequent　fault　processing　technique.　Through　analysing　the　features　of　short　currents,　an　early　detection　method　for　short-circuit　faults　in　10-kV　middle-voltage　distribution　networks　based　on　the　Mallet　decomposition　algorithm　is　proposed.　The　reliability　of　the　proposed　method　under　some　abnormal　non-fault　conditions　is　studied　as　well.　Various　types　of　short-circuit　faults　in　distribution　systems　are　simulated　in　the　MATLAB/SIMULINK　environment.　The　wavelet　detection　method　(WDM)　and　other　methods　are　implemented　to　realize　fast　detection　of　faults,　and　the　detection　results　are　analysed　and　compared.　In　addition,　the　experiments　for　different　faults　are　accomplished　in　a　physical　simulation　system　and　current　waveforms　are　sampled.　The　algorithm　programs　of　the　WDM　are　written　in　a　hardware　detection　system　to　achieve　real-time　fault　detection.　The　simulation　and　test　results　show　that　the　proposed　early　detection　method　can　realize　millisecond-level　fault　detection,　adapt　different　fault　conditions　effectively,　and　has　sufficiently　high　calculation　efficiency　in　hardware.