Many proteins suspected of causing neurodegenerative diseases exist in diverse assembly states. For most, it is unclear whether shifts from one state to another would be helpful or harmful. We used mutagenesis to change the assembly state of Alzheimer disease (AD)-associated amyloid-beta (Abeta) peptides. In vitro, the "Arctic" mutation (AbetaE22G) accelerated Abeta fibrillization but decreased the abundance of nonfibrillar Abeta assemblies, compared with wild-type Abeta. In human amyloid precursor protein (hAPP) transgenic mice carrying mutations adjacent to Abeta that increase Abeta production, addition of the Arctic mutation markedly enhanced the formation of neuritic amyloid plaques but reduced the relative abundance of a specific nonfibrillar Abeta assembly (Abeta*56). Mice overexpressing Arctic mutant or wild-type Abeta had similar behavioral and neuronal deficits when they were matched for Abeta*56 levels but had vastly different plaque loads. Thus, Abeta*56 is a likelier determinant of functional deficits in hAPP mice than fibrillar Abeta deposits. Therapeutic interventions that reduce Abeta fibrils at the cost of augmenting nonfibrillar Abeta assemblies could be harmful.