PURPOSE: To identify anatomic determinants of reading performance in eyes with foveal-sparing geographic atrophy (GA). DESIGN: Prospectively recruited, cross-sectional study, SIGHT (clinicaltrials.gov identifier, NCT02332343). PARTICIPANTS: Patients with foveal-sparing GA secondary to age-related macular degeneration (AMD). METHODS: Monocular best-corrected visual acuity and reading acuity together with reading speed were assessed using Radner charts. Fundus autofluorescence, near-infrared reflectance, and spectral-domain OCT images were acquired using a Spectralis device. The minimal required reading rectangle (M3R), 19 letters $times$ 2.4 lines in the smallest readable print size of an individual eye, was computed. The status of the M3R was determined as either free of atrophy or involved in the atrophic process, and the impact on reading was assessed. MAIN OUTCOME MEASURES: Radner reading score (logRAD) and reading speed (words per minute [wpm]). RESULTS: A total of 45 eyes of 31 patients (30 women; mean age, 76.14 years [range, 64.17-89.22 years]) were included. Median best-corrected visual acuity was 0.20 logarithm of the minimum angle of resolution (logMAR; Snellen equivalent, 20/32). Reading score was 0.52 logRAD (IQR, 0.30-1.4 logRAD) and maximum reading speed was 141.19 wpm (IQR, 105.52-164.62 wpm). In 27 eyes, the M3R was involved in the atrophic process. This was associated with a significant worsening in Radner score (1.21 logRAD [IQR, 0.46-1.40 logRAD] vs. 0.31 logRAD [IQR, 0.20-0.51 logRAD]; P < 0.001) and reading speed (110.84 wpm [IQR, 90.0-131.92 wpm] vs. 162.34 wpm [IQR, 137.51-176.66 wpm]; P = 0.002). Eyes in which the M3R was nonatrophic additionally showed an increase in reading speed with decreasing print size (peak increase, +73.08 wpm [IQR, 27.43-86.64 wpm] compared with the largest test sentence). CONCLUSIONS: The results indicate that a defined area on the retina that can be assessed by retinal imaging is required for unhindered reading in patients with foveal-sparing GA. The findings highlight that smaller test sentences can be read faster by patients with this AMD subphenotype. Our results allow prediction of reading impairment based on imaging parameters in clinical routine and may support establishing anatomic surrogate end points in clinical trials. Furthermore, the findings could be used to facilitate the adjustment of magnifying reading aids.