In this paper, based on the content addressing process of associative memory, the optimization method of binary associative model is proposed by using a neural network. The error tolerance of the optimum model is analyzed statistically. It is shown that the performance of the optimum model is improved.