Thenoisedatainverticalcomponentrecordsof85seismicstationsinFujianProvinceduring2012isusedastheresearchobjectinthispaper.Thenoisedataispidedintofiveminutesegmentstocalculatethepowerspectra.Thehighreferencelineandlowreferencelineofstationarethenidentifiedbydrawingaprobabilitydensityfunctiongraph(PDF)usingthepowerspectralprobabilitydensityfunction.Moreover,accordingtotheanomaliesofPDFgraphsin85seismicstations,theabnormalnoiseispidedintofourcategories:droppedpacket,lownoise,highnoise,andmediannoiseanomalies.Afterwards,fourselectionmethodsarefoundbythehighorlownoisereferencelineofthestations,andthesystemofreal-timemonitoringofseismicnoiseisformedbycombiningthefourselectionmethods.Noiserecordsof85seismicstationsinFujianProvinceinJuly2013areselectedforverification,andtheresultsshowthattheanomalousnoise-recognitionsystemcouldreacha90%successrateatmoststationsandtheeffectofselectionareverygood.Therefore,itcouldbeappliedtotheseismicnoisereal-timemonitoringinstations.