Progress in the structure determination of bacteriorhodopsin, the protein component of purple membrane from Halobacterium halobium has been limited by the lack of three-dimensional phase information between 6 and 3 Å resolution. By analogy with X-ray methods, it is possible that heavy-atom labelling of the membrane crystal may provide heavy-atom derivatives that can be used for phasing by the multiple isomorphous replacement method. This paper describes the screening of heavy-atom compounds as potential derivatives, and the evaluation of the data collected from these heavy-atom-labelled membranes. Improvements in the methods for collecting electron diffraction data and analysing and merging the data are presented. Diffraction patterns of purple membrane samples were taken at -120°C to minimize radiation damage. About 30 heavy-atom compounds were tested for use as potential derivatives. The diffraction patterns from labelled membranes were analysed by examining 6·5 Å difference Fourier maps. Two heavy-atom compounds were selected for three-dimensional data collection at 3 Å resolution. In addition, a full set of native data at -120°C was collected to 2·7 Å resolution. The intensity merging, heavy-atom derivative evaluation, heavy-atom refinement and the calculation of phases are presented. Phases are compared to those determined by electron microscope imaging, and limitations of the method are discussed. It is concluded that, with the present accuracy of data collection and the present magnitude of ΔF/F available for the derivatives, the phasing power is too small. The phases that are obtained are not sufficiently accurate to provide a reliably interpretable map. It may be possible, however, to use the heavy-atom derivative data in difference Fourier calculations in which the presence of a peak would confirm the phases calculated from a model or obtained by electron microscope imaging. © 1990 Academic Press Limited.