Face recognition systems have gained more attention during the last decades. Accurate features are the corner stones in these systems where the performance of recognition and classification processes mainly depends on these features. In this chapter, a new method is proposed for a highly accurate face recognition system. Exact Gaussian-Hermit moments (EGHMs) are used to extract the features of face images where the higher order EGHMs are able to capture the higher-order nonlinear features of these images. The rotation, scaling and translation invariants of EGHMs are used to overcome the geometric distortions. The non-negative matrix factorization (NMF) is a popular image representation method that is able to avoid the drawbacks of principle component analysis (PCA) and independent component analysis (ICA) methods and is able to maintain the image variations. The NMF is used to classify the extracted features. The proposed method is assessed using three face datasets, the ORL, Ncku and UMIST which have different characteristics. The experimental results illustrate the high accuracy of the proposed method against other methods.