The discovery of sequence-specific gene silencing by endogenous double-stranded RNAs (dsRNA) has propelled synthetic short-interfering RNAs (siRNAs) to the forefront of targeted pharmaceutical engineering. The first clinical trials utilized 21-nucleotide (nt) siRNAs for the treatment of neovascular age-related macular degeneration (AMD). Surprisingly, these compounds were not formulated for cell permeation, which is required for bona fide RNA interference (RNAi). We showed that these "naked" siRNAs suppress neovascularization in mice not via RNAi but via sequence-independent activation of cell surface Toll-like receptor-3 (TLR3). Here, we demonstrate that noninternalized siRNAs induce retinal degeneration in mice by activating surface TLR3 on retinal pigmented epithelial cells. Cholesterol conjugated siRNAs capable of cell permeation and triggering RNAi also induce the same phenotype. Retinal degeneration was not observed after treatment with siRNAs shorter than 21-nts. Other cytosolic dsRNA sensors are not critical to this response. TLR3 activation triggers caspase-3-mediated apoptotic death of the retinal pigment epithelium (RPE) via nuclear translocation of interferon regulatory factor-3. While this unexpected adverse effect of siRNAs has implications for future clinical trials, these findings also introduce a new preclinical model of geographic atrophy (GA), a late stage of dry AMD that causes blindness in millions worldwide.