The basic theories of cuckoo search (CS) and Hopfield Neural Network (HNN) are introduced, and the application of Hopfield Network in the digit recognition is researched. Aiming at the problem that Hopfield Neural Network can easily fall into local minimum, a new method that Hopfield network combines CS is presented. The method uses the global search capability of CS for complex, multimodal, nonlinear and non-differentiable functions to make Hopfield network keep a higher success rate even if noise-to-Signal ratio is high, and a simulation was carried out. Experiment results show that this method has a better performance.