Alzheimer disease (AD) is an incurable, progressive neurodegenerative disorder with a long presymptomatic period that is clinically characterized by cognitive and behavioural impairment, social and occupational dysfunction and, ultimately, death. In the USA, AD was the sixth most common cause of death in 2015 and showed the largest age-adjusted increase (16%) relative to 2014 (ref. 1). According to current estimates, 17% of people aged 75-84 years in the USA have AD, and the disease costs the country US$236 billion per year. The prevalence is projected to triple by 2050 to >15 million, with annual costs of >$700 billion 2. Diagnosis of AD is usually based on medical history and clinical findings, sometimes corroborated by brain imaging. Therapies are symptomatic and do not affect disease progression; currently, cholinesterase inhibitors and the N-methyl-d-aspartate receptor antagonist memantine are the only available options. Despite extensive research into the pathophysiology of AD, the large number of drugs entering clinical development and the enormous expenditure on large and complex trials, no new drug has been approved since memantine in 2003. Many reasons have been proposed to explain this failure, including inappropriate patient selection, variable rates of progression, suboptimal dosing, drug exposure and/or target engagement, inappropriate time of intervention, inappropriate outcome measures and low sensitivity of clinical scales. In addition, an incomplete understanding of AD pathophysiology might have led to selection of the wrong targets. Most of the drugs tested for AD in the past 20 years have targeted the accumulation of the amyloid-β (Aβ) peptide. In this article, we consider the current anti-Aβ drugs and the possible reasons for their failure to provide meaningful clinical benefits. Anti-Aβ therapies have been extensively tested in sporadic late-onset forms of AD, which will be the main focus of our discussion; no therapeutic studies have yet been conducted in individuals with the less common familial forms. The amyloid cascade hypothesis The Aβ peptide is generated by metabolism of amyloid precursor protein (APP), a type I transmembrane glycoprotein of 695-770 amino acids. APP is cleaved close to the membrane by an extracellular protease known as α-secretase. This cleavage liberates a soluble extracellular fragment, sAPPα. APP is also cleaved by an aspartyl protease known as β-secretase 1 (BACE1),