Specific amino acid sequence segments have been assigned to locations in the structural map of bacteriorhodopsin using two-dimensional neutron diffraction data and a model building analysis. Models are constructed computationally by building specific regions of the amino acid sequence as alpha helices and then positioning the helices on axes indicated by the density map of Henderson and Unwin (Nature [Lond.]. 1975, 257:28-32). Neutron diffraction data were collected from samples of stacked, oriented "native" purple membranes as well as purple membranes containing different kinds of deuterated amino acids. Models differing in the assignments of helices to specific axes and in rotations of the helices about those axes were tested against the neutron data using a weighted residual factor to rank the models. This residual factor was calculated between observed and predicted intensity differences for pairs of data sets. Using this approach, a small set of related models has been found that predicts the observed intensity changes between five independent data sets. These models are inconsistent with the proposed locations of the retinal chromophore and the carboxyl terminus and with any of the previously proposed models for bacteriorhodopsin.