In order to extract vibration signals of rotating machinery, a new method combining wavelet packet demodulation with the full vector spectrum is explored to extract fault features based on the information fusion technique. First, vibration data is sampled synchronously for multichannel information from the same section of a rotor. Then the two sets of original data from every sensor are preprocessed using the wavelet transformation to remove noise interference. The recomposed signals are merged using the full vector spectrum. Lastly, the fusion data is analyzed with the envelope spectrum. Analysis of the simulation and experimental results shows that when the wavelet packet envelope is applied to the two sets of vibration signals, their characteristic frequencies and spectrum energy distribution are obviously different. To obtain clear and all sided fault information, full information demodulation is introduced based on the wavelet packet. The merged data is analyzed using the new method, resulting in clear characteristic frequencies. Thus, the proposed method based on data fusion shows a great advantage compared with the envelope demodulation.