A novel combination of theoretical approaches was exploited to predict which amino acid residues of various botulinum neurotoxin serotypes participate in forming ion channels. Estimates of sequence hydrophobic moments were used initially to identify residues within amphipathic regions in the N-terminal half of the heavy chain. A neural network algorithm was then used to make additional secondary structural predictions for these regions. Together, these approaches predicted a complimentary pattern of four, adjacent amphipathic, possibly transmembrane, regions that may be separated by solvent-exposed loops. Both the hydrophobic moment and the neural net analyses predicted that at least one of these amphipathic segments may be in an extended conformation. These theoretical results are discussed in view of our current knowledge of other transmembrane structures.