This study established a method of detecting temporomandibular joint sounds based on signal-to-noise ratios. After comparing the temporomandibular joint signals obtained from three different sites over the skin, the articular eminence was found to be the best site for detecting temporomandibular joint sounds; this site provided the highest mean amplitude in the time domain waveform. However, using an electret condenser microphone at the intra-auditory meatus provided a broader, 20-decibel signal-to-noise bandwidth, which resulted in minimized artifacts. This method may be useful for recording temporomandibular joint sounds in the differential diagnosis of various temporomandibular joint conditions using spectral analysis.