Homology modelling is now a commonly used method for predicting the 3D structure of a protein for which the amino acid sequence is known, providing that at least one suitable "template" 3D structure - based on sequence homology or predicted fold - is available. Stereochemical restraints, derived both from standard values and the "template" structures, are usually incorporated into the model building process. The accuracy of the resulting model structure(s) can be enhanced if, alongside these, additional restraints are used based, e.g., on NMR data, biochemical data and/or theoretical data. This talk will illustrate how distance restraints derived in these ways can be used with the program MODELLER1, presenting several examples from our recent work. Firstly, the use of distance restraints derived from NMR experiments in conjunction with homology modelling will be illustrated by several cytochrome P450-substrate interactions2-4. Secondly, the use of theoretically-derived distance restraints will be illustrated by the modelling of a PDZ domain-tripeptide complex5. Thirdly, the combined use of biochemically- and theoretically-derived distance restraints will be illustrated by the modelling of ligand interactions with a glutamate receptor6. The talk will also illustrate software we have developed recently for identifying, from a family of models, the "most representative" model7, and the "core atoms" and domain(s)8.