For　single-phase　grounding　faults　in　resonant　earthed　systems,　the　similarity　between　the　transient　zero-sequence　current　signals　of　the　fault　line　and　healthy　lines　is　lower　than　that　between　two　healthy　lines.　A　novel　approach　based　on　the　time-frequency　spectrum　similarity　of　the　transient　fault　zero-sequence　current　signals　was　presented　to　detect　the　fault　line　in　distribution　network.　After　extracting　a　certain　window　data　of　transient　fault　zero-sequence　current,　all　frequency　band　coefficients　for　each　line　could　be　got　through　wavelet　packet　decomposition　and　reconstruction.　Then　they　were　divided　into　small　time-frequency　pieces　which　had　the　same　frequency　bandwidth　and　time　interval.　The　time-frequency　spectrum　diagram　and　matrix　of　all　lines　could　be　obtained　by　calculating　the　amplitude　of　time-frequency　pieces.　Regarding　the　time-frequency　spectrum　matrix　as　the　pixel　matrix　of　digital　image,　the　method　for　image　similarity　recognition　combined　with　comprehensive　similarity　coefficient　was　used　for　fault　line　detection.　Some　factors　that　might　influence　the　effect　of　fault　line　detection　in　the　actual　application,　such　as　noise,　two-point　grounding,　different　compensation　degrees,　arc-grounding　fault,　and　asynchronous　sampling　etc,　were　taken　into　account.　Simulation　results　verified　the　robustness,　accuracy　and　reliability　of　the　proposed　approach　for　fault　line　detection.　©　2013　Chinese　Society　for　Electrical　Engineering.