Image alignment is an absolute requirement for three-dimensional (3-D) reconstruction from serial sections, and Fourier correlation is the most powerful way to compute alignments. The rotational and translational components of misalignment can be corrected by an iterative correlation procedure, but for images having significant differences, alignment can fail with a likelihood proportional to the extent of the differences. We found that translational correction was determined much more reliably when low-pass filters were applied to the product transforms from which the correlations were calculated. Rotational corrections based on polar analyses of the auto-correlations of the images instead of on the images directly contributed to more accurate alignments. These methods were used to generate 3-D reconstructions of brain capillary modules from serial-section mosaics of digitized transmission electron micrographs.