The reliability and validity of a frequency discrimination (FD) task were tested in 16 people with specific language impairment (SLI) and 16 people with normal spoken language (controls). The FD thresholds of the 2 groups indicated that FD thresholds for 25-ms and 250-ms tones were remarkably stable across 18 months. The FD thresholds were lower for control listeners than for listeners with SLI for both duration conditions, and the FD thresholds for both groups of listeners were lower for 250-ms tones than for 25-ms tones. Moreover, the FD thresholds were influenced little by nonperceptual, task-related abilities (e.g., paired-associative learning, memory for temporal order, sustained attention, and control of attention) of the listener groups. The significant group difference between the mean FD thresholds of the SLI and control groups was explained by a subgroup of people with SLI who had particularly poor thresholds compared with those of controls and the majority of the SLI group. This subgroup did not differ from the remainder of the SLI sample in terms of age or nonverbal ability but was characterized by very poor reading that was associated with poor phonemic awareness.