In this paper, we present a method for time-frequency analysis and classification of temporomandibular joint (TMJ) sounds based on evolutionary spectrum. Many researchers have worked on sounds recorded from patients with pain and/or mechanical disfunction at TMJ. It is generally accepted that a detailed analysis of TMJ sounds might offer Valuable information for diagnosis and initiation of a treatment. In this work, TMJ sounds from patients with different symptoms were recorded by means of accelerometers during jaw opening and closing cycles. Than, four main categories of TMJ signals were defined based on patient's medical examination and evolutionary spectra. A method is presented to classify TMJ sounds using their joint time-frequency moments and neural networks. (C) 2000 The Franklin Institute, Published by Elsevier Science Ltd. All rights reserved.