Although the assumption of an overall globular scattering entity can be useful for determining crystallographic phases for a protein at low resolution, there is a point where this pseudoatomic model must be abandoned for further phase refinement. Using 6 A resolution electron diffraction data from aquaporin (AQP-CHIP) as an example, phases of the 16 most intense reflections from a previous direct solution (Dorset & Jap (1998). Acta Cryst. D54, 615-621) were modified with a Hadamard error-correcting code to produce potential maps very similar to the ones obtained using phases from the Fourier transform of averaged electron micrographs. The choice of the optimal phase set was made via the cross correlation of experimental with anticipated density histograms using the autocorrelation function of the latter histogram as the desired endpoint.