An ophthalmic fundus camera was modified for spectral imaging of the human retina. Spectral channels were separated using 30 narrow bandpass interference filters in the visual range from 400 to 700 nm. A monochrome digital charge-coupled device camera was used as a detector. Using this system, spectral fundus images were recorded from 72 voluntary human subjects: 55 diabetic patients and 17 healthy control subjects. Furthermore, using the gathered spectral data, optimal illuminations' spectral power distributions for the detection of different lesions of diabetic retinopathy and retinal landmarks (e.g., macula) were calculated. Optimized illuminations were found to enhance the visibility and contrast of retinal features. Computational example images are presented.