We present a new approach for determining dynamical domains in large proteins, either based on a comparison of different experimental structures, or on a simplified normal mode calculation for a single conformation. In a first step, a deformation measure is evaluated for all residues in the protein; a high deformation indicates highly flexible interdomain regions. The sufficiently rigid parts of the protein are then classified into rigid domains and low-deformation interdomain regions on the basis of their global motion. We demonstrate the techniques on three proteins: citrate synthase, which has been the subject of earlier domain analyses, HIV-1 reverse transcriptase, which has a rather complex domain structure, and aspartate transcarbamylase as an example of a very large protein. These examples show that the comparison of conformations and the normal mode analysis lead to essentially the same domain identification, except for cases where the experimental conformations differ by the presence of a large ligand, such as a DNA strand. Normal mode analysis has the advantage of requiring only one experimental structure and of providing a more detailed picture of domain movements, e.g. the splitting of domains into subdomains at higher frequencies. Proteins 1999;34:369–382. © 1999 Wiley-Liss, Inc.