Three-dimensional structures of proteins, experimental or predicted, show us how these molecular machines actually work. With the help of information on disease-related mutations, they can also show us how they malfunction in diseases. Such understanding, currently lacking for most human diseases, is an important first step before designing drugs or therapies to cure specific diseases. Here we used homology modeling to model human disease-related proteins, and studied structural characteristics of disease related mutations and compared them with non synonymous SNPs. 1484 domains from 874 proteins were modeled, and together with experimentally determined structures of 369 domains they provided the structural coverage of 48% of total residues in 1237 human disease proteins. We found that disease-related mutations have statistically significantly preference to form clusters on protein surfaces. In contrast, the non-synonymous SNPs appear to be randomly distributed on the surface. We interpret these results as an indication that disease mutations affect protein-protein interaction interfaces. This interpretation is supported by the analysis of 8 experimentally determined complexes between disease proteins, where disease-related mutations are clearly located in the binding interface of proteins, while SNPs are not. The non-uniform distribution of disease mutations indicates that we can use this feature as guidance in modeling and evaluating human disease proteins and their complexes. We set up a resource for Disease Protein Models (DPM at http://ffas.burnham.org/DPM), which can be used for studying the relation between disease and mutation/polymorphism sites in the context of protein 3D structures and complexes.