The effectiveness of removing salt and pepper noise lies on the noise detection and the intensity estimation of noisy pixel, but existing filters show no very good performance for them. In view of this, we propose a decision filter based on image texture features. The proposed method performs noise detection by taking full advantage of the characteristics of salt and pepper noise, i.e., the intensity of noise takes extreme values and the noises are independent of noise free pixels, as well as the texture features of image; thus, correctly discriminates the noisy pixels from the noise free pixels. And based on the characteristic of smoothly varying intensities of image texture, the proposed method groups the neighbor noise free pixels; and then, in the light of correlation and significance of probability, takes the median of the group, which is closest to the mean of neighbor noise free pixels, as the intensity of noisy pixel. The experimental results show that, compared to existing filters, the proposed method can perform a more accurate noise detection; and its resulted PSNR increases averagely by more than 1.9dB, IEF increases averagely by more than 119. Thus, it can conclude that the denoising performance of the proposed method is superior to the existing filters.