To avoid the adverse effects of edge burrs and the difference of contrast ratio, a corner detection algorithm via multi-scale differential mode similarity is proposed. Based on the edge contour extraction, it extracts the local differential mode at each edge pixel by using the Anisotropic Gaussian Directional Derivative (ANDD) filters. Considering the fact that the differential mode at single pixel suffers from edge burrs, thus the similarity between the differential modes at a pixel and its support region is constructed as corner measure to enhance the stability of the corner measure. Finally, the measures at three scales are fused into a new corner measure to enhance the noise robustness and to improve the localization accuracy of corner measure. The experimental results show that the proposed algorithm behaves better performance in detection accuracy ratio, false detection ratio, localization and repeatability compared with three state-of-art algorithms.