Abstract: A new texture feature called direction measure is proposed in this paper. It is based on the principle that in a homogeneous texture the gray values of pixels along any direction must change according to a certain regularity (deterministic or stochastic). Direction measure has many different forms such as first-order direction measure, second-order direction measure and high-order direction measure. These different forms of direction measure can extract firstorder, second-order, and high-order statistical properties of image textures. Usually, the highorder direction measure is more effective than other direction measures, and has a very high classification rate for many kinds of textures. The derivation of direction measures only needs additions, substractions and multiplications. So less computational requirement is another advantage of direction measures.