Listeners' performance in identifying voiced and unvoiced vowels, masked by noise or other vowels, is compared with predictions by a model of auditory pitch analysis and sound segregation. The listening experiments show that the subjects use pitch to separate the vowels from the background: Identification of unvoiced vowels is consistently poorer in these conditions than that of voiced vowels. When the background is also voiced (simultaneous vowels), identification scores rise with an increasing difference between the pitches of the two vowels. The model clearly outperforms the listeners for unvoiced vowels. Predictions for voived vowels are closer to human performance. However, in that case the typical dependency of identification scores on the pitch difference cannot be observed. Implications for further development of the model are discussed.