Aiming　at　the　problem　of　Gaussian　noise　interference　and　order　in　the　process　of　low-frequency　oscillation　of　wide　area　measurement　system,　we　propose　a　new　method　based　on　adjacent　coefficient　tunable　Q-factor　wavelet　transform　(TQWT)　and　improved　TLS-ESPRIT　algorithm　to　identify　the　modes　of　low-frequency　oscillation　signal　in　power　grid.　In　the　proposed　method,　the　TQWT　is　used　to　decompose　the　power　signal　to　obtain　the　initial　wavelet　coefficients,　and　the　adjacent　coefficient　threshold　rule　is　used　to　deal　with　the　wavelet　coefficients,　and　reconstruct　the　processed　wavelet　coefficients　using　inverse　TQWT;　then　an　improved　TLS-ESPRIT　algorithm　is　utilized　to　identify　the　low-frequency　oscillation　modes　parameters.　The　results　of　numerical　simulations,　the　IEEE　four-machine　two-area　simulations,　and　the　actual　case　simulations　of　North　American　power　grid　show　that　the　proposed　method　can　accurately　identify　low-frequency　oscillation　modes　parameters,　and　has　better　anti-noise　performance　and　higher　fitting　accuracy　than　other　methods.　The　proposed　method　has　strong　practicability,　and　can　realize　on-line　identification　better,　which　will　lay　the　foundation　for　the　further　research　of　low-frequency　oscillation　suppression.　©　2019,　High　Voltage　Engineering　Editorial　Department　of　CEPRI.　All　right　reserved.