Three-dimensional electron-microscopic structural analysis requires the combination of many different tilted views of the same specimen. The relative difficulty of tilting the sample to high angles >60 degrees without introducing severe distortion due to different focal distances across the specimen entails that the observable range of electron diffraction data is often limited to this range of angles. Thus, it is generally not possible to observe the diffraction maxima that lie within the conical region of reciprocal space around the direction perpendicular to the electron microscope grid. The absence of data in this region leads to a predictable distortion in the object, and for +/-60 degrees tilting makes the resolution essentially twice as bad in the direction perpendicular to the grid as it is for the in-plane image. Constrained density map modification and refinement methods can significantly reduce these effects. A method has been developed, tested on model cases, and applied to the electron-microscopic structure determination of bacteriorhodopsin in order to visualize the location of linking regions between helices. Electron-microscopic structural analysis of bacteriorhodopsin (Henderson and Unwin. 1975 Nature [Lond.] 257:28-32.) showed that the molecule consists of seven rods of density each nearly spanning the lipid bilayer. Owing to the distortion introduced by the missing conical region of reciprocal space data, no density was visible for the polypeptide segments linking the alpha-helices. Density in the refined maps indicates the location of at least five of the extrahelical segments of the polypeptide. The total number of possible ways of interconnecting the helices is reduced from 7! (5,040) to the five most consistent possibilities without recourse to other considerations. In addition, the density for the helical regions is more uniform and cylindrical throughout their length, and the length of the helices increases from 35 to 45 A, close to the membrane thickness of 49 A obtained for membranes dried in vacuo. Only three of the five structures consistent with the location of observed linkers place the seventh helix, onto which the chromophore can be attached by reduction in the light, at a position consistent with the main peak for deuterated retinal in the structure, as derived from neutron diffraction analysis. Two of these models are also consistent with the possible location of some of the reduced chromophore on helix B, at lys 40/41 after reduction in the dark, as well as lys 216 on helix G.