BACKGROUND: To examine the association of diabetes and diabetic retinopathy (DR) with retinal ganglion cell (RGC) loss. DESIGN: Observational case-control study. PARTICIPANTS: Type 2 diabetes cases and age-gender matched controls without diabetes. METHODS: Spectral-domain optical coherence tomography (OCT) parameters of RGCs were calculated after automated segmentation of macular scans. DR severity was graded on fundus photographs using the modified Airlie House Classification system. Generalized estimating equation was used to compare OCT parameters between cases and controls, adjusted for covariates. MAIN OUTCOME MEASURES: Average ganglion cell-inner plexiform layer (GC-IPL) and average retinal nerve fibre layer (RNFL) thicknesses. RESULTS: We analyzed 227 cases and 227 controls. The mean age (years) of cases was 58.3 and controls was 58.1 (P = 0.13). Among cases, 101 had none, 25 had mild and 101 had moderate or severe DR. Compared with controls, GC-IPL and RNFL were thinner in all cases [mean difference (95% confidence interval [CI]): GC-IPL -4.49 µm (-2.92; -6.06), RNFL -0.93 µm (-0.09; -1.85)], including cases with no DR [mean difference (95% CI), GC-IPL -4.37 µm (-2.72; -6.02), RNFL -1.06 µm (-0.10; -2.02)]. Cases with any DR had thinner GC-IPL than controls [mean difference (95% CI): GC-IPL -4.81 µm (-2.12; -7.50)]. Among cases, subjects with moderate or severe DR had thinner GC-IPL than subjects with no DR [mean difference (95% CI): GC-IPL -2.07 µm (-0.08; -4.07)]. CONCLUSIONS: RGC loss is present in subjects with diabetes and no DR, and is progressive in moderate or severe DR. RGC neuronal damage in diabetes and DR can be clinically detected using OCT.