In cases where the structure of a single protein is represented by an ensemble of conformations, there is often a need to determine the common features and to choose a 'representative' conformation. Such cases occur, for example, with structures determined by NMR spectroscopy, analysis of the trajectory from a molecular dynamics simulation, or an ensemble of structures produced by comparative modelling. Presented here are fully automatic methods for (1) defining the atoms in an ensemble with low spatial variance across that ensemble (i.e. the 'core' atoms) and the domains in which these atoms lie, and (2) clustering an ensemble into conformationally-related sub-families, involving a novel clustering technique applicable to any data set. These methods have been incorporated into the C programs NMRCORE and NMRCLUST respectively. The latter is the first fully automated clustering technique and the results from it have been incorporated into IDITIS, the protein tertiary structure database from Oxford Molecular Ltd. Both methods have been integrated into a free available World Wide Web server called OLDERADO.;This (1) contains an automatically generated database of representative structures, core atoms and domains determined for the 449 ensembles of NMR-derived structures in the Protein Data Bank (PDB) in May 1997, and (2) allows the user to upload a PDB formatted file containing the co-ordinates of an ensemble of structures. OLDERADO is integrated into the PDB "3DB Browser" search engine permitting widespread access by the protein community. The server returns in real time information on the residues comprising domains, the structures that comprise each conformational sub-family and an interactive Java based 3D viewer to visualise the domains and clusters. Such information is useful for example for selecting conformations to be used in comparative modelling and parts of structures to be used in molecular replacement.