In this paper, we propose a bimodal distribution of random neuronal activity threshold with different probabilities, to consider the influences on the retrieval properties of neural network. It is shown that the system successfully retrieves information even if the number of stored patterns exceeds the critical value of the pure Hopfield model.