PURPOSE: Lipofuscin contained in the retinal pigment epithelium (RPE) is the main source of fundus autofluorescence (FAF), the target of an imaging method useful for estimating the progression of geographic atrophy (GA) in clinical trials. To establish a cellular basis for hyperfluorescent GA border zones, histologic autofluorescence (HAF) was measured at defined stages of RPE pathologic progression. DESIGN: Experimental study. PARTICIPANTS AND CONTROLS: Ten GA donor eyes (mean age ± standard deviation, 87.1 ± 4.0 years) and 3 age-matched control eyes (mean age ± standard deviation, 84.0 ± 7.2 years) without GA. METHODS: The 10-micrometer-thick sections were divided into zones of RPE morphologic features according to an 8-point scale. Any HAF excited by 488 nm light was imaged by laser confocal microscopy. The HAF intensity summed along vertical lines perpendicular to Bruch's membrane at 0.2-μm intervals served as a surrogate for FAF. Intensity profiles in 151 zones were normalized to grade 0 at a standard reference location in each eye. Cross-sectional area, mean, and sum autofluorescence for individual RPE cells were measured (cellular autofluorescence [CAF]). MAIN OUTCOME MEASURES: Statistically significant differences in intensity and localization of HAF and CAF at defined stages of RPE morphologic progression for GA and control eyes. RESULTS: The RPE morphologic features were most abnormal (cell rounding, sloughing, and layering; grade 2) and HAF intensity profiles were highest and most variable immediately adjacent to atrophic areas. Peaks in HAF intensity frequently were associated with vertically superimposed cells. The HAF value that optimally separated reactive RPE was 0.66 standard deviations more than the mean for uninvolved RPE and was associated with a sensitivity of 75.8% and a specificity of 76.3%. When variable cell area was accounted for, neither mean nor sum CAF differed significantly among the RPE pathologic grades. CONCLUSIONS: Areas with advanced RPE alterations are most likely to exhibit clinically recognizable patterns of elevated FAF around GA, but may not predict cells about to die, because of vertically superimposed cells and cellular fragments. These data do not support a role for lipofuscin-related cell death and call into question the rationale of treatments targeting lipofuscin.