PURPOSE: To recognize different patterns of distribution signals from a difference in grade by abruptly varying the grade in a proximity area when the different distribution signals to be recognized on a feature segmentation line are close to each other.

CONSTITUTION: When the distribution signal 4 of a pattern B is given, the grade of the pattern B is obtained by comparative collation with a membership function(MF) 3 and the grade of a pattern A is obtained by comparative collation with the MF 2 of the pattern A. For example, the distribution signal 4 of the character '3' of the pattern A and the character '5' of the pattern B on the feature segmentation line 1 are close. Then the MF2 and MF3 which abruptly vary in the proximity area are set. When the distribution signal 4 of the character '5' is given, a high grade is obtained by comparative collation with the MF2 of the character '5' and a low grade is obtained by comparative collation with the MF3 of the character '3'. Consequently, it can be recognized from the difference in grade between the characters that the given distribution signal 4 corresponds to the character '5'.