This paper examines the role of component phase and level on vowel identification and interprets the results in terms of the shapes of the waveforms occurring at the outputs of the filters in a nonlinear basilar-membrane model. Four normally hearing subjects were asked to identify which of six possible vowel-like harmonic complexes was presented on each trial. The stimuli were complex tones containing the first 35 harmonics of a 100-Hz fundamental. All of the harmonics below 3000 Hz were equal in amplitude except for three pairs of successive harmonics, at frequencies corresponding to the first three formants of six vowels, which were incremented in level relative to the background harmonics by 1, 2, 4, 8, and 16 dB. The components in the harmonic complexes were added in four different starting phase relationships; cosine, random, Schroeder positive, and Schroeder negative. The stimuli were presented at three overall levels; 85, 65, and 45 dB SPL. Performance was similar for the random and Schroeder-negative phases and did not vary as a function of level. Performance for the cosine- and Schroeder-positive-phase conditions was better than for the other two phase conditions, but decreased as the level was reduced. Performance for all four phase conditions was equivalent for the lowest level. The variation in performance as a function of level and component phase is explained in terms of the shapes of the temporal waveforms that would occur at the output of nonlinear "basilar-membrane filters" [H. W. Strube, J. Acoust. Soc. Am. 79, 1511-1518 (1986)], with asymmetric phase responses about the center frequency.