The classification of ships is one of the most important problem in the preliminary design stage. Reasonable classification must be provided with the wide range of application as well as the good correlations between variables which usually conflict each other. For this purpose, it is necessary to make clear the similarity or dissimilarity between ship types. A discriminator for the ship type is proposed in this paper. 20 characteristic values of ships are converted into the principal components, then these values are put into the discriminator to specify the ship type. The ability of the discriminator can be improved interactively by learning process. More than a half of ships in our data file has been discriminated properly. The remaining ships which failed in discrimination are useful to show the relations of the ship clusters in principal particulars space.