Abstract: This paper analysis the feature extraction method of handwritten Chinese character. The contour direction feature(CDF)and directional distance distributions feature(DDDF) are extracted from Chinese character image as a pair of feature vectors having good complementarity. After dimensions reduction of original feature using Karhunen-Loeve transform, these two feature vectors are combined to produce a new feature vector has high discriminating powers. Basing on the characteristic of RBFNN classifier, a novel architecture which integrates feature fusion and modular RBF neural networks classifier approaches into a small set handwritten Chinese character recognition system is presented. Experiments show that the system has achieved impressive performance and the results are informing.