This paper proposes a figure character recognition method based on features aiming at the complexity of character image,a large quantitity of input data and the slow convergence speed in recognizing and researching character image.Thirteen construction features designed and extracted in this paper compose a thirteen-dimension hyperplane space,wherein a character recognition system model is developed using the neural network cluster method(10 cluster number),the MATLAB with powerful matrix operating ability and the efficient code of C Language.Experiment results indicate that the recognition time and rate are still satisfactory under the indistinctive feature differences.