It is known that Karhunen-Loeve orthogonal system can be applied to unsupervised clustering. And a method for unsupervised clustering by the K-L system were proposed and some results were given by the author. However, this method has two weak points as follow. (1) The algorithm of this method is complex. (2) Therefore, it is difficult for us to apply this method to multi-category problem. Then, first the principle of the method for unsupervised clustering by K-L system are explaned. Secondly, in order to get rid of these weak points, a explanation is given to a new method for unsupervised clustering by the use of the procedure which adds a pattern in one class to other class, where a part of this method has already shown in reference 〔8〕. Finally, computer-simulated experiments are carried into effect in order to test usefullness of the present method.