Single-variable　criterion　methods　of　arc　fault　diagnosis　are　greatly　influenced　by　uncertain　factors　and　difficult　to　extract　the　characteristic　quantities,　aiming　at　which,　a　multi-variable　criterion　based　on　EMD(Empirical　Mode　Decomposition)　and　PNN(Probabilistic　Neural　Network)　is　proposed.　Time-frequency　decomposition　of　arc　current　is　carried　out　by　EMD　analysis　method,　and　the　fault　characteristic　signal　is　extracted　by　signal　correlation　theory　automatically.　The　set　of　multi-variable　characteristic　vectors　is　formed　by　analyzing　the　dimensionless　index　of　fault　characteristic　signals.　On　this　basis,　an　arc　fault　diagnosis　model　based　on　PNN　is　established.　The　accuracy　of　the　proposed　model　is　verified　by　analyzing　current　waveforms　of　kettles,　vacuum　cleaners,　halogen　lamps,　drills,　fluorescent　lamps　and　computers　before　and　after　arcing.　Results　show　that　the　proposed　method　solves　the　problems　of　difficult　feature　extraction　and　cross-repetition　in　single-variable　criterion　fault　diagnosis,　and　its　accurate　rate　is　over　90%.　©　2019,　Electric　Power　Automation　Equipment　Press.　All　right　reserved.