Methods for determining hearing aid settings often incorporate measurements of most comfortable loudness (MCL) and uncomfortable loudness (UCL) levels. This study examined the variability of loudness measures and their correlation to threshold data, using speech stimuli presented to hearing-impaired subjects. MCLs, UCLs, speech reception, and speech detection thresholds were obtained from 50 subjects having sensorineural impairments. The stimuli were CID W-2 spondees spoken by three female clinicians. Three MCLs and UCLs were obtained within each session, using ascending runs and a closed-set response list. Fifteen subjects were retested twice over intervals ranging from a week to several months. Between-session variability for the loudness measurements was less than or equal to 10 dB across sessions and speakers for the majority of subjects, with a tendency for the MCL and UCL to increase slightly over time. Significant variability was attributed to the use of live-voice presentation by different clinicians. High positive correlation was found between threshold and loudness data for subjects with relatively flat audiometric configurations but not for subjects demonstrating sharply sloping hearing losses.