In order to construct a reference model of human-machine identification texture direction,a metric algorithm for global texture direction is proposed.Firstly,a Gabor filter is designed to reduce the component less important direction and enhance that of global direction for description texture.Then, the histogram of local edge probability is constructed to describe the distribution of global texture and to give a quantitative value of measurement. Finally, the proposed algorithm is applied to comparing the accuracy of two experimental data sets from Tamura Texture Model and Brodatz texture database. Results indicate that the accuracy of the algorithm is well consistent with that of human visual perception measurement,and it is higher 10% than that of other algorithms.The algorithm can satisfy the requirements of a better reference model of human-machine identification texture direction for the higher precision.