Research into Alzheimer's disease (AD) has been guided by the view that deposits of fibrillar amyloid-beta peptide (Abeta) are neurotoxic and are largely responsible for the neurodegeneration that accompanies the disease. This 'amyloid hypothesis' has claimed support from a wide range of molecular, genetic and animal studies. We critically review these observations and highlight inconsistencies between the predictions of the amyloid hypothesis and the published data. We show that the data provide equal support for a 'bioflocculant hypothesis', which posits that Abeta is normally produced to bind neurotoxic solutes (such as metal ions), while the precipitation of Abeta into plaques may be an efficient means of presenting these toxins to phagocytes. We conclude that if the deposition of Abeta represents a physiological response to injury then therapeutic treatments aimed at reducing the availability of Abeta may hasten the disease process and associated cognitive decline in AD.