The growth of intelligibility of speech stimuli as a function of level above hearing threshold can be computed from the “circuit parameters” of the hearing mechanism by applying Shannon's concepts of channel capacity, equivocation, and “bits.” In the ear, the unit of response is an effective “least count,” derived from experimental data on hearing by means of the equations for a model resembling a frequency‐selective circuit [E. Corliss, J. Acoust. Soc. Amer.41, 1500–1516 (1967)]. The model predicts that the number of least counts available rises as the one‐fourth power of the signal intensity above threshold. Experimentally, this growth rate is observed for the intensity‐resolving power of the ear. Approximately the same power law is observed for the sensation of loudness. The model ascribes both effects to the same mechanism. From the observed integration time of the ear, the model predicts the rate at which transitions of single counts can be detected. From the counting rate and the integration time, the channel capacity available at the ear and its increase with level above threshold can be computed. The information content of speech as a source function is evaluated from the rate at which single “distinctive features” of speech phonemes are produced. Intelligibility scores can be predicted from the ratio between the rate at which information is being produced by the source and the rate at which the receptor can accept the source material. The scores predicted agree fairly closely with experimental data on random‐word and random‐syllable intelligibilities. This agreement shows that the listener need recognize no more than a single distinctive feature of each phoneme to display the recognition functions that have been observed. From a theorem of C. Shannon [Inform. and Control1, 6–25 (1957)] relating code length and error probability, one can show that the channel capacity required for polysyllabic words is lower than the channel capacity required for monosyllabic words because the duration of correlated utterance may be taken as a code length. Evidently, contextual effects are not prominent in the intelligibility of random‐word lists; the hearing process involved is primarily recognition of groups of sounds; meaning is secondary. The results also lead to the inference that a direct relation may exist between channel capacity and perceived loudness when speech is transmitted over a broad‐band system, and suggest that loudness functions for impaired ears might prove to be correlated with intelligibility functions.