This paper proposes a new Enhanced Nearest Feature Space Classifier (ENFS) which inherits the generalization capability from Nearest Feature Space method. Additionally, estimated variance can optimize the class seperability in the sense of Bayes error, and has improve the classification power in reduced PCA subspace. Gabor wavelets representation of face images is an effective approach for both facial action recognition and face identification. Perform PCA dimensionality reduction on the downsampled Gabor Wavelets features can be effectively for face recognition. In our experiments, ENFS with proposed Gabor Wavelets Features shows very good performance, which can achieve 98.5% maximum correct recognition rate on ORL data set without any preprocessing step.