The topological structure of soil pores determined the ability of soil moisture retention and conductivity, which had a significant impact on soil ecological processes. However, the existing pore identification methods had the problems of low pore identification accuracy and low operational efficiency. In order to solve the problems, a fast fuzzy C-means (GFFCM) method based on the grayscale-gradient features of soil CT images for pore identification was proposed. The grayscale-gradient two-dimensional feature matrix was established by Laplace operator to describe the characteristics of pore boundary. Combined with soil prior knowledge, the initial membership matrix was constructed and the number of clusters was estimated. Then, based on the determined initial conditions, the traditional fuzzy C-means was used to realize the fuzzy division of soil structure. Finally, the fuzzy clustering result was optimized with the GFFCM method by pore identification standard to accurately identify the soil pore structure. The methods were applied to the soil CT images with unsaturated state and compared with the traditional FCM method and the fast FCM method (FFCM), the GFFCM method had the lowest identification error rate and the smallest number of iterations, which indicated that the GFFCM method had the highest recognition accuracy. Besides, the method could overcome the shortcomings of the traditional FCM method in initializing the membership matrix and number of clusters, so it solved the problem that the initial value influenced the identification accuracy and had the advantage of high computational efficiency.