Natural language can be used as a knowledge representation language, but direct manipulation of it in a machine is cumbersome. Natural language, however, can be converted into other knowledge representation language which is easier to handle in a computer. A way of automatic knowledge base construction by using Korean text is described in this article. In the proposed method, each input sentence is translated into a dependency structure according to the dependency relation defined in the parser. The dependency structure is used to generate a corresponding conceptual graph and it is stored in a knowledge base which can retrieve a graph efficiently.