Thresholds and consonant recognition were measured in six low-pass maskers as a function of masker bandwidth for hearing-impaired subjects and for normal-hearing subjects listening in spectrally shaped broadband noise (SSBB). SSBB was adjusted such that thresholds in that masker for a normal-hearing listener were equal to a hearing-impaired listener's absolute thresholds. Thresholds measured in low-pass maskers were higher for hearing-impaired than for normal-hearing subjects for signal frequencies both within and outside masker passbands, although threshold differences were larger for signal frequencies outside masker passbands. Slopes of functions relating consonant recognition to speech level were not significantly different between groups, due to the presence of SSBB for the normal-hearing listeners. However, 25% of observed scores for hearing-impaired listeners, compared to only 5% of observed scores for normal-hearing listeners, were significantly poorer than predicted by the articulation index (AI), when AIs were computed using subjects' absolute thresholds. Better correspondence between observed and predicted scores in low-pass maskers was achieved when AIs were derived empirically from thresholds measured in each low-pass masker. Hence poorer-than-predicted consonant recognition scores in low-pass maskers were accounted for by higher-than-normal thresholds in those maskers.