Combination tones (CT's) produced by two-tone stimuli (f1 and f2) at relatively low sound levels contradict the classical view that auditory mechanics is an essentially linear process that suffers significant percentage distortion only at high sound levels. CT level and phase behavior were measured extensively with pitch-cancellation and loudness-balancing methods. Relative level of the most prominant CT 2f1-f2 was nearly unaffected by stimulus level but decreased sharply with increasing frequency interval, being typically 15%–20% for f2/f1 = 1.10. In contrast, the difference tone was audible only for stimulus levels above 50 dB sensation level (SL), grew to 15% relative distortion only for estimated stimulus levels exceeding 100 dB SL, and was relatively insensitive to frequency interval. Other CT's of the form f1-n (f2-f1) were heard, and these decreased sharply in level with increasing integer n. The fact that CT's above the stimulus frequencies were inaudible is not caused by stimulus masking but rather reflects instead a mechanical frequency selectivity in the nonlinear source. Thus, although auditory mechanical analysis not essentially linear, the concept that the cochlea performs a limited resolution frequency-place transformation is supported. Physical studies of the cochlea should elucidate the nature of the hypothesized essential cochlear mechanical nonlinearity.