Ethan Merritt ®rst encountered crystallography as a graduate student in computer science, and was delighted to have found a research ®eld offering a wide scope for developing and applying computational techniques in pursuit of biological goals. He earned a PhD in the group of M. Sundaralingam in Madison, dividing his efforts between macromolecular graphics, conformational modeling of nucleotides and the use of anomalous scattering to determine the stereospeci®city of metal/nucleotide coordination in ATP-binding enzymes. He then joined the research staff of the Stanford Synchrotron Radiation Laboratory, where he worked to implement the X-ray optics, beamline-control systems and data-processing software needed to turn MAD phasing from a promising idea into a routine technique for structure determination. This effort culminated in a series of collaborations, notably with Hans Freeman and Wayne Hendrickson, which yielded the ®rst protein structure determinations entirely from MAD phasing. In 1989, he moved to the University of Washington Medical School in Seattle. There his research interests have included structure-based drug design, development of the Raster3D graphics package and the improved exploitation of synchrotron radiation in protein crystallography. The review presented here grew from a convergence of these interests.# 1999 International Union of Crystallography Printed in Denmark ± all rights reserved Recent technological improvements in crystallographic data collection have led to a surge in the number of protein structures being determined at atomic or near-atomic resolution. At this resolution, structural models can be expanded to include anisotropic displacement parameters (ADPs) for individual atoms. New protocols and new tools are needed to re®ne, analyze and validate such models optimally. One such tool, PARVATI, has been used to examine all protein structures (peptide chains >50 residues) for which expanded models including ADPs are available from the Protein Data Bank. The distribution of anisotropy within each of these re®ned models is broadly similar across the entire set of structures, with a mean anisotropy A in the range 0.4±0.5. This is a signi®cant departure from a purely isotropic model and explains why the inclusion of ADPs yields a substantial improvement in the crystallographic residuals R and R free . The observed distribution of anisotropy may prove useful in the validation of very high resolution structures. A more complete understanding of this distribution may also allow the development of improved protein structural models, even at lower resolution.