In　this　paper,　a　two-stage　mode　identification　algorithm　including　preprocessing　and　identification　steps　is　introduced　to　solve　the　problems　of　measuring　noise　and　inaccurate　mode　identification　in　the　analysis　of　low-frequency　oscillation　(LFO)　in　a　wide　area　measurement　systems　(WAMSs).　S-G　filter　de-noising　is　utilized　for　the　preprocessing.　An　adaptive　total　least　squares-estimation　of　signal　parameters　via　rotational　invariance　techniques　(TLS-ESPRIT)　algorithm　with　the　order　setting　of　the　singular　value　accumulation　percentage　adjacency　increment　ratio　(SVAPAIR)　is　utilized　for　the　identifying　stage.　The　S-G　filter　is　used　to　mitigate　the　measuring　of　LFO　noise　of　the　system.　Furthermore,　the　SVAPAIR　is　adopted　to　achieve　an　adaptive　and　precise　determination　of　the　LFO　order,　and　then　LFO　parameters　are　extracted　by　the　adaptive　TLS-ESPRIT　algorithm.　The　proposed　algorithm　has　been　tested　and　verified　using　the　IEEE　four-generator　two-area　system　and　the　actual　measured　data　of　the　LFO　accident　in　the　North　American　power　grid.　The　simulation　and　experimental　result　showed　that　the　proposed　algorithm　has　better　adaptability,　anti-noise　characteristic,　and　robustness　compared　to　those　of　the　conventional　Prony,　matrix　pencil　(MP),　and　TLS-ESPRIT　algorithms.