The motivation of the work is to develop a signal processing methodology for noninvasive diagnosis of knee osteoarthritis in an early stage. The sound signal that is emitted from knee when it moves is called Vibroathrographic (VAG) signal. Analysis of this sound signal will help in diagnosis of the knee joint problems. In this project a model based approach for sementing the VAG signals, followed by feature extraction and classification is proposed. This could be used to get some indication whether the signal is from a normal knee or from an abnormal knee. The proposed scheme also has the capability for finding the depth of severity of the damage and it can also localize the angle range of the knee swing, where the damage has occurred. As a result, the project gave an accuracy of 70.4% with leave-one-out method. After doing the classification using the segments, finally it has been calculated how many segments from each signal has been correctly identified. A total of 30 knee sound signals from normal and abmoraml knees has been used in this work and out of that 26 signals has been classified properly (either normal or abnormal) and 4 signals got misclassified with a successful classification accuracy of 86.7%.