In view of the problems of feature extraction in face recognition, a new face image feature extraction and recognition method—Gabor Discriminant Locality Preserving Projections (GDLPP) is proposed in this paper. GDLPP first gets the high-order statistical information by calculating the Gabor wavelet representation of face images. Based on LPP, GDLPP takes into account the inter-class information, changes the objective function, and extracts the discriminant feature of face for recognition. The proposed method was tested and evaluated in the USPS database, Yale face database and AR face database. Nearest neighborhood algorithm was used to construct classifiers. The experimental results show that GDLPP has good performance even if pose, illumination, face expression and train sample number change.