A Korean character is made up of sequential additions of basic alphabets under positioning rules, thus there are a large amount of characters and consequently there exist many similar patterns. Accordingly, it is an important problem to avoid the ambiguities among the similar characters. Recently a syntactic method which extracts basic alphabets sequentially according to positioning rules under the control of tree grammars is applied, but the ambiguity problem remains, and it has been doubtful whether the designed grammar is effective in case of hand-written characters. In this thesis, a recognition system of Korean characters composing of two main bodies, description and structural analysis, is suggested in order to minimize the ambiguities. Detection of a number of nodes over the input pattern and codification of strokes between nodes by some predetermined primitives is presented in order to describe the input character as a bi-directional graph. The structural analysis contains two procedures, one is top-down nondeterministic parsing and the other is bottom-up segmentation process to resolve the fatal ambiguity problems occurring from the concatenations between basic alphabets. Very high recognition ratio has obtained through the experimentation by digital computer, and the author have made sure that this system is powerful for recognizing Korean characters.