Both meaningful (sense) and meaningless (nonsense) syllables of the consonant-vowel-consonant type (CVC syllables) and short sentences consisting of 8 or 9 syllables were presented in quiet and in noise to 20 young subjects with normal hearing and to three groups of 20 subjects each with presbycusis, with Menière's disease and with noise-induced hearing loss. All materials were uttered by a female speaker. The masking noise consisted of continuous noise shaped in accordance with the long-term average spectrum of the speaker. For each individual, the level of the noise was chosen halfway between the speech reception threshold (SRT) for sentences in quiet and 100 dBA. For all groups of subjects in quiet, the SRT for whole-sentence correct scores (sentence SRT) corresponded closely to the SRT for phoneme scores with sense CVC syllables in quiet (CVC phoneme SRT). Averaged across all groups of subjects, sentence SRT in quiet could be predicted within 4.2 dB from CVC phoneme SRT in quiet and sentence SRT in noise within 1.8 dB from CVC phoneme SRT in noise. The prediction error for sentence SRT in quiet using the pure-tone average (PTA) of 0.5, 1 and 2 kHz was 6.0 dB; for sentence SRT in noise using the PTA of 2 and 4 kHz, it was 2.1 dB. In view of the smaller measurement error, a direct measurement of sentence SRT in noise is advisable.