The effects of spectral shaping on speech recognition were investigated for hearing-impaired listeners with flat and steep audiometric configuration. Three frequency responses were tested: uniform frequency gain, high pass filtering, and a response shaped relative to each subject's loudness discomfort level curve. Speech-recognition performance was measured at four levels (from 80 to 95 dB SPL) using nonsense syllable (NST) and synthetic sentence (SSI) tests, presented against a background of "cafeteria noise." No significant differences in performance on the NST were observed between the two subject groups across all spectral shapes (frequency response) and presentation levels. On the SSI, performance of subjects with flat audiometric configuration was highest using the uniform frequency response, while performance of listeners with steep configuration was poorest for the uniform response. The recognition data were compared with predictions of relative performance using a modification of the Articulation index (AI). The AIs provided accurate estimates of relative performance across spectral shapes but were not consistent with relative performance as a function of presentation level. The results indicate that the selection of spectral shape for optimal performance is influenced by the particular speech task used to test recognition and also suggest that, with further validation, the AI may provide an objective technique for selecting optimal spectral shape.