Recent progress in the field of electron cryo-microscopy and image analysis has shown that there is an overwhelming need to interpret medium resolution (5 to 10 A) three-dimensional maps. Traditional methods of fitting amino acid residues into electron density using molecular modeling programs must be supplemented with further analysis. We have used a potential of mean force (PMF) method, derived from Boltzmann statistics in protein structure, to generate models for the packing of alpha-helices, using pairwise potentials between amino acid residues. The approach was tested using the three-dimensional map of a recombinant cardiac gap junction membrane channel provided by electron cryo-crystallography (Unger et al., 1997; 1999a, 1999b) which had a resolution of 7.5 A in the membrane plane and 21 A in the vertical direction. The dodecameric channel was formed by the end-to-end docking of two hexamers, each of which displayed 24 rods of density in the membrane interior, which was consistent with an alpha-helical conformation for the four transmembrane domains of each connexin subunit. Based on the three-dimensional map and the amino acid sequence for the 4 transmembrane domains determined by hydropathy analysis, we used the modeling utility SymServ (Macke et al., 1998) to build hexameric connexons with 24 transmembrane alpha-helices. Canonical alpha-helices were aligned to the axes of the rods of density and translated along the density so that the center of masses coincided. The PMF function was used to evaluate 162,000 conformations for each of the 24 possible alpha-helical packing models. Since the different packing models yielded different energy distributions, the pair potential function appears to be a promising tool for evaluating the packing of alpha-helices in membrane proteins. The analysis will be refined by energy calculations based on the expectations that the outer boundary of the channel will be formed by hydrophobic residues in contact with the lipids.