Abstract: After analyzing Fourth Order Cumulant (FOC) of harmonic signals, we divide the FOC into three parts. One is the FOC of the cubic frequency (phase) coupling components of the signals. The second is the FOC of the double frequency (phase) coupling components of the signals, which is proposed by us first. The last is the FOC of the rest components of the signals. On the bases of the study, the FOC diagonal slice is used to extract the cubic frequency (phase) coupling feature, double frequency (phase) coupling feature and the "sub-band energy" feature of ship-radiated noise. In terms of the features, the three type ships are classified by means of artificial neural network. The correct classification rates of A, B and C ships are 92.5%, 92.7%, 88.6%, respectively. The results shows the method is effective and practical.