Abstarct: Underwater radiated noise of ships in the sea has some information about their machineries . The information can be used to recognize them . It is a difficult task to recognize ships by underwater radiated sound due to there are a lot of changes in sea`s environment , special characteristics of sound transmission channels in underwater and interference-wave in sea’s environment. One of the most important approach in ships recognition by radiated sound underwater is extracting individual features from the underwater sound of ships which is an attracting subject for researchers. The main target of this thesis is to suggest an algorithm for ships recognition by their underwater radiated sound. In this thesis we use adding spectrums respectively that are obtained from segmentation of input signal and form a binary image to extract features. Extracted features are given to the input of neural network . Then the kind of ship will be determined. Different experiments have been done on a databaxse of sound ships. This databaxse is consist of the sound of 20 files of heavy ship, 10 files of boat, 60 files of average ship, 20 files of submarine,12 files of dolphin. The correct recognition rate in 7 classes is % 97.18 with SNR=10dB and it is %96.26 in 5 classes with the same SNR .