The original facial features can be mainly divided into two categories, texture characteristic and structure characteristic. Both of them have advantages and disadvantages. This paper proposes a new feature, which is fusion of texture and structure feature. A standard fiducial point detector is applied to locate five facial key points. The new feature will be generated adaptively by revising the gray value of pixel according to the distance to its relevant key point. The new feature involves geometric structure feature and information of key points, which is more robust than the original texture feature. Meanwhile, the five key points are beneficial to part-based face recognition. Experimental results show that the new feature can obtain 5% higher recognition rates than original texture features generally, and the improved part-based method, which combines the new feature and the face blocking, achieves recognition rates close to 100% in two different available databases.