See Mercado and Hetz (doi:10.1093/brain/awx107) for a scientific commentary on this article.Signalling through the PERK/eIF2α-P branch of the unfolded protein response plays a critical role in controlling protein synthesis rates in cells. This pathway is overactivated in brains of patients with Alzheimer’s disease and related disorders and has recently emerged as a promising therapeutic target for these currently untreatable conditions. Thus, in mouse models of neurodegenerative disease, prolonged overactivation of PERK/eIF2α-P signalling causes sustained attenuation of protein synthesis, leading to memory impairment and neuronal loss. Re-establishing translation rates by inhibition of eIF2α-P activity, genetically or pharmacologically, restores memory and prevents neurodegeneration and extends survival. However, the experimental compounds used preclinically are unsuitable for use in humans, due to associated toxicity or poor pharmacokinetic properties. To discover compounds that have anti-eIF2α-P activity suitable for clinical use, we performed phenotypic screens on a NINDS small molecule library of 1040 drugs. We identified two compounds, trazodone hydrochloride and dibenzoylmethane, which reversed eIF2α-P-mediated translational attenuation in vitro and in vivo. Both drugs were markedly neuroprotective in two mouse models of neurodegeneration, using clinically relevant doses over a prolonged period of time, without systemic toxicity. Thus, in prion-diseased mice, both trazodone and dibenzoylmethane treatment restored memory deficits, abrogated development of neurological signs, prevented neurodegeneration and significantly prolonged survival. In tauopathy-frontotemporal dementia mice, both drugs were neuroprotective, rescued memory deficits and reduced hippocampal atrophy. Further, trazodone reduced p-tau burden. These compounds therefore represent potential new disease-modifying treatments for dementia. Trazodone in particular, a licensed drug, should now be tested in clinical trials in patients.