Experimental data and model predictions on amplitude-modulation detection are presented. The sensorineural hearing-impaired listeners had a flat hearing loss in the stimulated frequency region and showed recruitment. Normal and impaired listeners showed the same TMTF for different noise carriers (fc=2 kHz; bandwidth=16 Hz, 200 Hz). The presentation level was equal in units of categorical loudness scaling. Corresponding simulations were performed with a modified version of the modulation filterbank model [Dau et al., J. Acoust. Soc. Am. 97, 3273(A) (1996)], in which an increased absolute threshold and a loss of compression were introduced. The model can account for the data, using the same parameters for the scaling of the modulation filters, indicating that frequency selectivity in modulation is comparable to that of normal-hearing listeners. In the case of random noise as the carrier, thresholds are determined by the inherent statistics of the carrier. Therefore no effect of reduced compression can be expected, because test modulation and inherent carrier modulation are treated the same way. Using deterministic carriers, thresholds are only determined by internal noise added after the compression stage in the model. Therefore, a further experiment with a sinusoidal carrier was performed to examine the role of compression and internal noise on amplitude-modulation discrimination.