We present a technique for estimating the density of the human macular pigment noninvasively that takes advantage of the autofluorescence of lipofuscin, which is normally present in the human retinal pigment epithelium. By measuring the intensity of fluorescence at 710 nm, where macular pigment has essentially zero absorption, and stimulating the fluorescence with two wavelengths, one well absorbed by macular pigment and the other minimally absorbed by macular pigment, we can make accurate single-pass measurements of the macular pigment density. We used the technique to measure macular pigment density in a group of 159 subjects with normal retinal status ranging in age between 15 and 80 years. Average macular pigment density was 0.48 +/- 0.16 density unit (D.U.) for a 2 degrees -diameter test field. We show that these estimates are highly correlated with reflectometric (mean: 0.23 +/- 0.07 D.U.) and psychophysical (mean: 0.37 +/- 0.26 D.U.; obtained by heterochromatic flicker photometry) estimates of macular pigment in the same subjects, despite the fact that systematic differences in the estimated density exist between techniques. Repeat measurements over both short- and long-time intervals indicate that the autofluorescence technique is reproducible: The mean absolute difference between estimates was less than 0.05 D.U., superior to the reproducibility obtained by reflectometry and flicker photometry. To understand the systematic differences between density estimates obtained from the different methods, we analyzed the underlying assumptions of each technique. Specifically, we looked at the effect of self-screening by visual pigment, the effect of changes in optical property of the deeper retinal layers, including the role of retinal pigmented epithelium melanin, and the role of secondary fluorophores and reflectors in the anterior layers of the retina.