We exploited the large number of known allosteric crystal structures to systematically characterize local conformational changes in allosteric proteins toward the goal of increasing the theoretical understanding of the structural basis of protein allostery (protein switching) on the atomic scale. We compiled a set of 51 pairs of known inactive and active allosteric protein structures from the Protein Data Bank. To capture changes in different degrees of freedom important to protein structure, we have measured changes in dihedral angles and Cartesian displacements for backbones and side chains and rearrangements in residue-residue contacts for each protein on a per-residue basis. To identify which of these measured changes are crystallographically significant, we have determined thresholds for each measure based on distributions of motions by that measure in control sets of multiple crystal structures not exhibiting allosteric motions. Several examples show that these automated calculations reveal functionally interesting pictures of local motions which corroborate many features previously observed manually by crystallographers. In addition, statistical analysis of the calculated motions shows that on average, 20 percent of residues differ significantly between the two crystal structures of an allosteric protein in addition to possible changes in dynamics. Allosteric motion is more probable in weakly constrained local structural environments like loops and solvent-exposed regions than in strongly constrained environments like helices, strands, and buried regions, which suggests that proteins use hydrogen-bonding and contact constraints to regulate where motions happen in their structures. Backbone and contact motions are correlated at separations of up to 20 residues in sequence space and up to 20 Å in Cartesian space, which is enough to mechanically connect allosteric and functional sites in many proteins. Together, these observations suggest prediction and design rules for allosteric proteins.