Abstract: In order to overcome the problem that the diagnosis model prediction accuracy is generally not high an-d the training time is too long in analog circuit fault diagnosis. Putting forward a new method of analog circuit fa-ult diagnosis based on wavelet packet and BAGRNN. The method chooses the generalized regression neural netw-ork (GRNN) which has more advantages than the BP neural network as a network model, obtaining the fault char-acteristics of circuit by wavelet packet transform, then building the BAGRNN model by using bat algorithm whic-h the global search ability is strong and the search speed is fast to optimize smoothing factor for GRNN, finally a-pplying the optimized BAGRNN model for fault identification and classification. The simulation results show tha-t the model of BAGRNN greatly reduces the sample training time and has high prediction accuracy compared wit-h other diagnostic methods, the average diagnostic accuracy can be 97.1875%.