In　this　paper,　we　applied　empirical　mode　decomposition　method　to　analyse　rotor　startup　signals,　which　are　non-stationary　and　contain　a　lot　of　additional　information　other　than　that　from　its　stationary　running　signals.　The　methodology　developed　in　this　paper　decomposes　the　original　startup　signals　into　intrinsic　oscillation　modes　or　intrinsic　modes　function　(IMFs).　Then,　we　obtained　rotating　frequency　components　for　Bode　diagrams　plot　by　corresponding　IMFs,　according　to　the　characteristics　of　rotor　system.　The　method　can　obtain　precise　critical　speed　without　complex　hardware　support.　The　low-frequency　components　were　extracted　from　these　IMFs　in　vertical　and　horizontal　directions.　Utilising　these　components,　we　constructed　a　drift　locus　of　rotor　revolution　centre,　which　provides　some　significant　information　to　fault　diagnosis　of　rotating　machinery.　Also,　we　proved　that　empirical　mode　decomposition　method　is　more　precise　than　Fourier　filter　for　the　extraction　of　low-frequency　component.　(c)　2004　Elsevier　Ltd.　All　rights　reserved.