An algorithm on feature frequency extraction and fusion,in which some ideas of adaptive resonance theory are referred to,is proposed.In this algorithm,multi-normalization for spectrum vectors is made and a nonlinear threshold function is inserted into this normalization operation to suppress the interference frequencies caused by noises.Simultaneously,some proper positive feedbacks are used to strengthen the feature frequencies,and the fusion computation of different spectrums is also embedded in,all these proessing measures form a closed iterative computation loop,by which the feature frequencies can be extracted,fused and strengthened from the multiple spectrums,while those interference frequencies can be greatly suppressed.Three kinds of vibration noise spectrums of the electromotor of an air-conditioner are processed by this algorithm.The results show that for each kind of noise,a clear and accurate featurespectrum is obtained from the multiple spectrums,and its effectiveness is much better than that of the average spectrum method.