Polymerization of the amyloid beta (Abeta) peptide into protease-resistant fibrils is a significant step in the pathogenesis of Alzheimer's disease. It has not been possible to obtain detailed structural information about this process with conventional techniques because the peptide has limited solubility and does not form crystals. In this work, we present experimental results leading to a molecular level model for fibril formation. Systematically selected Abeta-fragments containing the Abeta16-20 sequence, previously shown essential for Abeta-Abeta binding, were incubated in a physiological buffer. Electron microscopy revealed that the shortest fibril-forming sequence was Abeta14-23. Substitutions in this decapeptide impaired fibril formation and deletion of the decapeptide from Abeta1-42 inhibited fibril formation completely. All studied peptides that formed fibrils also formed stable dimers and/or tetramers. Molecular modeling of Abeta14-23 oligomers in an antiparallel beta-sheet conformation displayed favorable hydrophobic interactions stabilized by salt bridges between all charged residues. We propose that this decapeptide sequence forms the core of Abeta-fibrils, with the hydrophobic C terminus folding over this core. The identification of this fundamental sequence and the implied molecular model could facilitate the design of potential inhibitors of amyloidogenesis.