When rotary machinery runs, vibration signals measured with sensors are mixed with all vibration sources and thus contain very strong noises. Mixed signals are difficult to separate using conventional methods of signal processing, such as in machine health monitoring and fault diagnosis. The principle and methods of blind source separation are introduced, and the blind source separation algorithm is indicated to be invalid in strong pulse noise environments. In these environments, vibration signals are first de-noised with the median filter method, and the de-noised signal is separated using the blind source separation algorithm. The results of the simulation verify the effectiveness of the proposed method. Finally, some mixed rotor vibration signals are dealed with using the proposed method. A new separation approach is thus provided for use on vibration signals in strong pulse noise environments.