Properly carried out, high-resolution X-ray diffraction data collection followed by careful least-squares refinement can give the spatial distribution of the high-frequency mean-square displacements in a protein. These displacements reflect both individual atomic fluctuations in hard variables (bond lengths and bond angles) and collective motions involving soft variables (torsion angles, nonbonded interactions). Lower frequency, large amplitude motions and rapid but improbable motions are not quantifiable, but they may lead to such complete disorder that their existence can at least be inferred from the absence of interpretable electron density for some sections of the structure. Interior residues are more rigid than groups on the surface, and structural constraints are reflected in restricted motion even for surface residues. Amplitudes of motion of 0.5 A or greater are not uncommon. The temperature dependence of these fast motions varies considerably over the structure. In general, large [chi 2] values have large temperature dependence, while small displacements are less affected by temperature; however, exceptions are common. Significant reduction in [chi 2] on cooling establishes that proteins are mobile even in the crystalline state, and that static disorder is not the dominant contributor to the individual mean square displacements. Disordered regions in electron density maps are no longer automatically taken as signs of errors in structure determination. It is now recognized that the absence of strong electron density is often an indicator of conformational flexibility. Some of the functional roles for protein dynamics are beginning to be understood. Missing from these results are the physicochemical details that can be extracted from thermal motion analysis of small molecule crystal structures. Application of these methods to protein data is very difficult, but it is well to remember that just over 10 years ago it was commonly felt that protein structures could not even be refined. Certainly some small, well-diffracting proteins should be amenable to many of the sophisticated small-molecule analyses, as they yield X-ray data to resolutions comparable to simple organic structures. The most important type of analysis that awaits is anisotropic B factor refinement, which would give the principal directions of motion added to the amplitude information now obtained. Unfortunately, refinement of unrestrained anisotropic thermal elipsoids requires six parameters for each atom instead of a single isotropic B parameter, and even 1.5 A resolution data do not provide enough overdeterminacy.(ABSTRACT TRUNCATED AT 400 WORDS)