We report a systematic study of small silver clusters, Ag n , Ag n + , and Ag n − , n = 1−7. We studied all possible isomers of clusters with n = 5−7. We tested 42 exchange−correlation functionals, and we assess these functionals for their accuracy in three respects: geometries (quantitative prediction of internuclear distances), structures (the nature of the lowest-energy structure, for example, whether it is planar or nonplanar), and energies. We find that the ingredients of exchange−correlation functionals are indicators of their success in predicting geometries and structures: local exchange−correlation functionals are generally better than hybrid functionals for geometries; functionals depending on kinetic energy density are the best for predicting the lowest-energy isomer correctly, especially for predicting two-dimensional to threedimenstional transitions correctly. The accuracy for energies is less sensitive to the ingredient list. Our findings could be useful for guiding the selection of methods for computational catalyst design.