Over the last several years, numerous biometrics algorithms have been investigated for face, fingerprints, iris, and voice reocognition applications. For face recognition research, projection-based face recognition system form the basis of numerous algorithms and studies . For biometrics identification and verification scenarios, we explicitly state the design decisions by introducing a generic modular face recognition system. We explored different implementations of preprocessing, feature extraction, and recognition module, and evaluate the different implementations using the FERET evaluation protocol. Our experiment includes changing the illumination normalization procedure, studying effects on algorithm performance of compressing images using JPEG and wavelet compression algorithms, and varying the number of eigenvectors in the representation. We perform series of experiments based on the standard FERET database and report results for identification and verification scenarios.