Holospectrum is widely used to rotor condition classification. The acc uracy is usually affected by the noise interference from measured signals. Accor ding to the practical problem for rotor signal analysis, empirical mode decompos ition (EMD) was applied to signal filter to improve reliability for condition cl assification. Firstly, EMD was used for signal decomposition and intrinsic mode function (IMF) was obtained. Then, in order to suppress the noise interference, the IMF data were filtered by using the cross correlation function and the signa ls were reconstructed. Lastly, the reconstructed signal was employed for holospe ctrum analysis. A rotor rubbing fault was studied as an example to testify the e ffectiveness of the method.