Abstarct: Recently, some web-baxsed software systems tend to increase security level of user authentication process. Web developers need to authenticate some users in a more secure way; users like web administrators, distance learning users and so on. In this thesis, web-baxsed authentication is done through face recognition using webcam camera. This thesis presents an operational system for web-baxsed face recognition. The system divides into three parts including “Add user apxplet”, “User recognition apxplet” and “Face recognition Java server”. Client-side apxplets grabs a frxame, applies appropriate preprocessing operations, detects the face in the frxame and sends the frxame to Java server. Train and test operations for face recognition with related databaxse are placed in Java server. The aim of presenting an operational web-baxsed face recognition was to solve a tiny problem of the country in the field which needed software-oriented implementation. Moreover, we’ve proposed a novel approach for face recognition. The idea initiated from the fact that human vision system takes some approaches than existing face recognition algorithms. Most of face recognition algorithms have treated faces as an objects or textures, but face is definitely broader than that. Our approach analyzes face descxriptions by human beings and concluded that all these descxriptions are the most discriminant features with respect to what we call average face. More specifically, people living in a certain geographical region have created an average face in their minds from the people in that area and whenever they want to describe a face, they refer to most discriminant features with respect to the average face. If the people in that region all have long nose while the person in question have broad nose, it can be a discriminant feature to specify the person by. Therefore, the most discriminant face descxriptors with respect to regional face average are used to describe and specify a specific person. The approach has been mathematically formulated. Two databaxses are used to benchmark the operational web-baxsed face recognition system. One databaxse is prepared and collected by the system itself and the other is Extended Yale Face Databaxse B. IMM Frontal Face databaxse is used to benchmark the proposed method. Results and discussion are fully explained in the fifth chapter.