A new texture feature extraction method using Local Walsh Transform (LWT) is presented. The definition of LWT is given. The statistical properties of LWT coefficients are analyzed. The texture discrimination performance of the moments of LWT coefficients are investigated. Detail examinations reveal that the LWT coefficients of the natural texture images usually do not yield to Gauss distribution, their even-order moments have high texture discrimination performance, while their odd-order moments have low texture discrimination performance. Hence, the even-order (2^nd, 4^th, 6^th order) moments of the LWT coefficients are selected as texture features. Compared with the other texture features defined by Haralick^［1］,Wang and He^［2,3］, Hui Yu^［5］, the texture features we present have the best texture discrimination performance.