Most of our understanding of chemistry derives from atomic-level structures obtained with single-crystal X-ray diffraction. Metal centers in X-ray structures of small organometallic or coordination complexes are often extremely well-defined, with errors in the positions on the order of 10(-4)-10(-5) A. Determining the metal coordination geometry to high accuracy is essential for understanding metal center reactivity, as even small structural changes can dramatically alter the metal activity. In contrast, the resolution of X-ray structures in proteins is limited typically to the order of 10(-1) angstrom. This resolution is often not sufficient to develop precise structure-activity relations for the metal sites in proteins, because the uncertainty in positions can cover all of the known ranges of bond lengths and bond angles for a given type of metal complex. Here we introduce a new approach that enables the determination of a high-definition structure of the active site of a metalloprotein from a powder sample, by combining magic-angle spinning (MAS) nuclear magnetic resonance (NMR) spectroscopy, tailored radio frequency (RF) irradiation schemes, and computational approaches. This allows us to overcome the "blind sphere" in paramagnetic proteins, and to observe and assign H-1, C-13, and N-15 resonances for the ligands directly coordinating the metal center. We illustrate the method by determining the bond lengths in the structure of the Co-II coordination sphere at the core of human superoxide dismutase 1 (SOD) with 0.7 pm precision. The coordination geometry of the resulting structure explains the nonreactive nature of the Co-II/Zn-II centers in these proteins, which allows them to play a purely structural role.