Many researchers have proposed that hearing aids should process sounds so as to restore loudness perception to 'normal'. We describe how a model for predicting loudness for people with cochlear hearing loss can be implemented in a digital hearing aid so as to calculate the frequency-dependent gains that would be required to achieve that goal. It is assumed that the input signal is processed using brief segments or 'frames'. For each frame, the spectrum is calculated, usually via a fast Fourier transform (FFT). From the spectrum, an excitation pattern is calculated for a normal car and for the impaired ear of the patient. The loudness model is then used to calculate the gain required at the centre frequency of each channel in the aid, so as to match the specific loudness in the normal and impaired ears. The whole process is repeated for each successive frame, with overlap of frames and with smoothing of the gain changes across frames. We describe both an 'exact' model, which prescribes a 'curvilinear' compression characteristic at each frequency, and an approximation using 'straight' compression, which is computationally less intensive. Limitations of the present approach are described, and the approach is compared with more traditional approaches using multichannel compression, and with previous approaches using loudness models for fitting hearing aids.