PURPOSE: To improve the probability of succeeding in recall, and to increase a storage capacity by learning a spurious state falling at the time of failing in recall as an exception pattern.

CONSTITUTION: A neural net is constituted of a neuron element 1, connecting part 5, connection coefficient setting means 7 by a storage pattern, and connecting coefficient setting means 9 by the exception pattern (pattern not to be recalled). At the time of operating the neural net, inputs I1-I5, and outputs O1-O5 are connected with the neuron element 1, and the connection coefficients are defined at the connecting part 5. Then, the recall information from the neuron element 1 is transmitted to the connection coefficient setting means 1 by the storage pattern, and the connection coefficient setting means 9 by the exception pattern, and an optimal learning is executed according to it, so that the learning information from both the connection coefficient setting means 7 by the storage pattern, and the connection coefficient setting means 9 by the exception pattern, can be transmitted to the connecting part 5.