In order to identify the fault of bearing through abnormal sound caused by the running bearing, an experimental platform for acoustic signal fault diagnosis of bearing is established. In the light of the poor signal-to-noise ratio, complex composition and weak fault feature of the acoustic signal, a microphone array signal feature extraction method of bearings is proposed based on sparse decomposition. The noise sources are identified and located through effective sound pressure fields and projections of holographic planes. Redundant dictionary is composed of coif4 wavelet and local cosine dictionaries. The impact features of sources′acoustic signal are extracted by sparse decomposition. The processing results of simulated and actual acoustic signals show that feature frequencies of fault acoustic signals under different rotating speed of bearing are accurately extracted by the proposed method, demonstrating the effectiveness and reliability of bearing fault identification through the microphone array signal.