PURPOSE: To provide a neural network simulator device prevented from interrupting processing even when there are many learning patterns.

CONSTITUTION: In the neural network simulator capable of applying functions of units respectively constituting an input layer, an intermediate layer and an output layer to respective processor elements(PEs) connected like a ring and executing learning processing using a back propagration method based upon data stored in a local memory(LM) connected to each PE, plural registers(Rs) are respectively connected to the PEs and a part of data necessary for the learning processing is inputted from the external through each R and applied to the corresponding PE.