Advances in the understanding of Alzheimer's disease (AD) suggest that pathogenesis is not directly related to plaque burden, but rather to soluble toxic amyloid-beta oligomers (AßO). Therapeutic antibodies targeting Aß monomers and/or plaque have shown limited efficacy and dose-limiting adverse events in clinical trials. These findings suggest that antibodies capable of selectively neutralizing toxic AßO may achieve improved efficacy and safety. To this end, we generated monoclonal antibodies against a conformational Aß epitope predicted by computational modeling to be presented on toxic AßO but not monomers or fibrils. The resulting lead antibody, PMN310, showed the desired AßO-selective binding profile. In vitro, PMN310 inhibited AßO propagation and toxicity. In vivo, PMN310 prevented AßO-induced loss of memory formation and reduced synaptic loss and inflammation. A humanized version (huPMN310) compared favorably to other Aß-directed antibodies showing a lack of adverse event-associated binding to Aß deposits in AD brains, and greater selective binding to AßO-enriched AD brain fractions that contain synaptotoxic Aß species. Systemic administration of huPMN310 in mice resulted in brain exposure and kinetics comparable to those of other therapeutic human monoclonal antibodies. Greater selectivity for AßO and the potential to safely administer high doses of huPMN310 are expected to result in enhanced safety and therapeutic potency.