Audibility-based approaches to hearing-aid selection generally have focused on the long-term average speech spectrum (LTASS). Advances in amplification technology (e.g., multiband signal processing, level-dependent frequency shaping, full dynamic range compression, adaptive compression) make it difficult to predict the audibility of short-term components of speech from the amplified LTASS. This study was designed to quantify the audibility of a specific phonemes as processed by two different hearing-aid circuits (linear and full dynamic range compression), and to investigate the relation between audibility and performance on a nonsense syllable recognition task. Data were obtained from three subjects with moderate sensorineural hearing loss. Nine unvoiced consonants were presented in two vowel contexts (/i/ and /a/) in both the pre- and post-vocalic position at three intensities. While the performance on selected conditions appeared to vary by hearing-aid type, only one subject showed a statistically significant difference between the two hearing-aid systems. Acoustic analyses revealed a variety of spectral and temporal changes to the speech signal following processing. Estimates of audibility were based upon each subject's thresholds and an acoustic analysis of the amplified signal that varied across phonemes and consonant position. A signal detection approach was used to predict performance from a simple measure of audibility.