A common way to understand structure in multimolecular systems is the coordination shell which comprises all the neighbors of an atom. Coordination, however, is nontrivial to determine because there is no obvious way to determine when atoms are neighbors. A common solution is to take all atoms within a cutoff at the first minimum of the radial distribution function, g(r). We show that such an approach cannot be consistently applied to model multicomponent systems, namely mixtures of atoms differing in size or charge. Coordination shells using the total g(r) are found to be too restrictive for atoms of different size while those using pairwise g(r)s are excessive for charged mixtures. The recently introduced relative angular distance algorithm, however, which defines coordination instantaneously from atomic positions, is consistently able to define coordination shells containing the expected neighboring atoms for all these systems. This more robust way to determine coordination should in turn make coordination a more robust way to understand structure. © 2017 Wiley Periodicals, Inc.