PURPOSE: To improve a recognition rate and the recognizing processing speed by deciding from the width of a pattern whether a separated pattern is an em or an en and combining a recognizing a rejected en pattern and other pattern.

CONSTITUTION: The device is provided width an input part 11, a CPU 12, a ROM 13m a RAM 14, a KANJI (Chinese character) dictionary 15, a non-KANJI dictionary 16 and output part 17. The CPU 12 detects the character width of the X direction, for example, pi (square form of Japanese syllabary) is discriminated to be the em, and when (g) is compared with the em mode W1, modes W2, W3 and W4 are smaller, therefore, it is discriminated to be an en mode. In case of 'setsu' (Chinese character), first, 'gen' (Chinese character) is segmented, a discriminating processing is executed, the collation of a non-kanji dictionary 16, in short, the distance calculation is executed and when the result of the obtained distance calculation is larger than a prescribed value, it is rejected, combined with an adjoining character area, a character area, which is 'setsu' (Chinese character), is formed and the processing to the em mode is proceeded.