Sounds are often produced by the temporomandibular joint (TMJ) during movement in both symptomatic and asymptomatic subjects. However, subjective methods of describing these sounds have been shown to have poor inter- and intra-observer reliability. In this study, a low cost system in which TMJ sounds were detected using the loudspeakers of lightweight in-ear headphones as microphones is evaluated. The sounds were recorded on tape and then analysed using a computer. Sounds were elicited by asking subjects to bring their teeth together with sufficient force to produce a tooth contact sound, then open their mouths as far as possible and then close again. Placing the microphones in the ears attenuated ambient sounds by 58%, thus providing a degree of immunity from ambient noise. Sampling was performed on the left microphone only at 3.4 kHz and from the left and right microphones together at 1.7 kHz for 60 TMJ sounds and 60 tooth contact sounds. Spectral analysis of sounds recorded at the two sample rates revealed no significant differences. Therefore, a sample rate of 1.7 kHz is adequate to resolve the frequency components present in the TMJ sounds. Although simply recording TMJ sounds does not give a direct measurement of the position of the mandible, using this protocol allows the length of the open close cycle to be determined. If the envelope of movement is assumed to approximate a sinusoid, then the direction of mandibular movement can be assumed to reverse at the half way point in the cycle. The accuracy of this assumption was calculated by comparing the mid-point of the cycle to the point of maximum gape in 129 cycles from nine subjects. The mean difference expressed as percentage of cycle length was 1.3 +/- 0.9%.