This paper proposed a new feature extraction method for Chinese character recognition by using optimized Gabor filters. Based on the theory of Gabor filters and the statistical information of Chinese character images, a simple but effective method to design Gabor filters was developed. Moreover, to improve the performances for low quality images, we modified the non-linear function used in previous research to regulate the outputs of Gabor filters adaptively. This paper also meliorated the feature extraction method to improve the discriminability of histogram features. Experiments had shown that our method perform excellently for images with noises, backgrounds or stroke distortions and can be applied to printed or handwritten character recognition tasks in low quality greyscale or binary images.