The secondary digital character recognition based on the character global feature and BP neural network is proposed in this paper. First of all, digital characters should be pre-recognized by the method of BP neural network, and a preliminary result can be obtained. After that, the secondary recognition to those characters which may be very close to confusing in the preliminary result should be arranged, so that, those digital characters confused can be distinguished and the better result can be reached. As a result, the method, combined with nonlinear and self-learning of the neural network, and invariance of the overall character feature of the shape and structure, can reach the high recognition accuracy on the condition of the low sample size.