Aiming to identify freshwater fish species automatically, passive acoustic signal samples of common freshwater fish were collected by the HTI-96-MIN standard hydrophone. A wiener filter and a sampling noise reduction method were used to preprocess the samples. Then frequency band energy of the samples was extracted by using wavelet packet decomposition algorithm. The layer number of the algorithm was four. The characteristic vectors of the samples were comprised of short-time average energy, short-time average zero-crossing rate, frequency band energy, and difference among the characteristic vectors of the four classes samples. Furthermore, a probabilistic neural network was used to identify freshwater fish species rapidly. As different values of the smoothing factor σ, the identification effect was studied. The results indicated that the identification accuracy was the highest when the smooth factor was between 9 and 19. The identification accuracies of ctenopharyngodon idellus, megalobrama amblycephala and crucian carp were all 100%. The identification accuracy of passive acoustic signals with zero fish was 77.3%. And the total accuracy was 94.3%. The proposed freshwater fish identification method using passive underwater acoustic signals had higher accuracy and less calculation. It provided a new way for identifying freshwater fish species quickly.