Abstract: One feature extraction method of modulation spectrum in the radiated noise of underwater targets was presented. Modulation curve is a linking route of maximum points of target waveshape. Maximum points of the target waveshape can be searched after applying the algorithm of wavelet transforms on the radiated noise. Then applying cubic spline technique, modulation curve are composed of these maximum points. Modulation spectrum is extracted from modulation curve. At last, the project is conducted by using fuzzy ART neural networks to the extracted character. The results show that it can efficiently classify underwater targets, at the same time it can show the period phenomenon of original wave.