Vibroarthrographic (VAG) signals, generated during the active flexion and extension of the leg, represent acoustic signals caused by joint vibration and can be used as useful indicators of the roughness, softening, breakdown, or the state of lubrication of the articular cartilage surface. This paper describes an efficient algorithm in order to improve the classification accuracy of the features obtained from the time-frequency distribution (TFD) of normal and abnormal VAG signals. VAG signals were correctly segmented by the dynamic time warping and the noise within the TFD of the segmented VAG signals was diminished by the singular value decomposition method. The classification of the knees as normal or abnormal was evaluated using a back-propagation neural network (BPNN). 1408 VAG segments (normal 1031, abnormal 377) were used for evaluating our devised algorithm by a BPNN and, consequently, the mean accuracy was 91.4plusmn1.7%. This algorithm could help to enhance the performance of the feature extraction and classification of VAG signal.