We performed an analysis of mutation patterns in all 10 hydrophobic regions (HRs) of presenilin-1 (PS-1) and PS-2 using a recent database of Alzheimer's disease (AD) mutations. The linear patterns were confirmed and extended to areas spanning as many as three faces of a given HR. The complementary areas of residues free of AD mutations were identified based on the location of non-pathogenic polymorphisms and PS-1 versus PS-2 amino acid discordances. Taking into account the location of areas of AD mutations and mutation-free areas/regions, we proposed a preliminary model of PS-1 structure using a general stick-out-mutation rule. To build a molecular structure of PS-1 and preserve features of the preliminary model, we used bacteriorhodopsin template in homology/comparative modelling. Two molecular models were built differing in the location of C-terminal fragment helices. The models properly distinguish residues belonging to AD-affected sites and non-pathogenic areas, and may be used for classification purposes. They also comply with experimental results, such as differences in accessibility of the catalytic residues in uncleaved PS-1, and binding of PEN-2 by the PS-1 NF motif.