Normal hearing subjects, listening to 2 conditions of low-pass, filtered speech, and 2 groups of listeners with sensorineural hearing loss, having audiometric roll-off frequencies analogous to those of the low-pass, filter cut-off frequencies (i.e., 900 and 2000 Hz), were tested with the California Consonant Test (CCT). Analysis or errors based on the three manners of articulation in the test stimuli--fricative, stop, and affricate--indicated that the probability of error per manner changes considerably across roll-off frequency. Because the distribution of test items on the CCT by manner of articulation is not consistent with that found in conversational speech, incorrect estimation of word recognition ability may result. It is suggested that the CCT be altered to reflect the distribution of the manner of articulation found in everyday speech.