PURPOSE: To obtain the neural network having versatility by executing learning while applying an evaluating value to a learning pattern in the case of learning the neural network, and varying a criterion corresponding to the quality of the learning pattern.

CONSTITUTION: This neural network executes learning by inputting a feature vector 1 to an input unit 2 and applying a category signal 4 and an evaluation value 5 of the learning pattern as teacher signals. A neural element 3 is equipped with an evaluation value corresponding to the evaluation vlue 5 of the learning pattern, and the evaluation value of the neural element is learnt so as to approximate the evaluation value of the learning pattern. In the case of recognition by the neural network, (k) pieces of neural elements having a small distance value are selected in a (k) neural element selecting part 7, and an expected value 9 of the evaluation value is calculated for each category. A judging unit 10 executes judgment processing or abortign processing of the category based on the maximum expected value of the evaluation value and the output value of a neural element 8 for aborting.