Traditional time-frequency analysis methods such as short-time Fourier transform (STFT) and Wigner-Ville distribution (WVD) cannot always work effectively for the complex rotor systems where the multiple faults are involved. A noise cancellation method for the rotor faults detection is proposed in this paper by combining the independent component analysis (ICA) scheme and the adaptive time-frequency analysis (ATFA) approach. In the proposed method, the raw picked data are first separated into different independent components (ICs) via the ICA according to the different vibration sources. Then the ICs are processed by the ATFA to obtain a clear vibration character for the fault diagnosis. Experiments on a rotor system with hybrid fault of the rotor imbalance and the bolt looseness are introduced to verify the feasibility and validity of the proposed scheme.