In restoration ecology, reference sites serve as models for areas to be restored and can provide a standard of comparison for restoration project outcomes. When reference sites are located a relatively long distance from associated restoration projects, differences in climate, disturbance history, and biogeography can increase beta diversity and may decrease the relevance of reference sites. Variation in factors at the scale of individual reference sites such as patch size, microclimate, barriers to dispersal, or soil chemistry can result in reference site species composition that is a nested subset of the regional species pool. In the western United States, restoration of riparian areas, particularly those occupied by Tamarix spp., has become a priority; however, little is known about suitable native replacement vegetation communities for relatively dry and saline riparian terraces that comprise many of the sites where Tamarix is removed prior to restoration activities. We studied plant communities on riparian terraces along five rivers in New Mexico, USA, to (1) determine whether the floristic composition of reference sites can be predicted by easily measured soil variables such as pH, salinity (electric conductivity), and texture; (2) examine the extent of distance decay in the compositional similarity of xeroriparian plant communities in the southwestern United States; and (3) determine the degree of nestedness in xeroriparian plant communities in relationship to soil variables. We found that sites clustered into groups based largely on variation in soil salinity and texture. Vegetation across all sites was highly nested with dominant, salt‐tolerant species found on most soil groups and salt‐intolerant subordinate species restricted to low‐salinity soils. The identity of subordinate species was largely site dependent, causing all sites to have the same low degree of similarity regardless of the distance between them. We conclude that, when planning restoration projects on dry and saline riparian sites, soil salinity and texture are good predictors of which species will be most suited to the area being restored, but a candidate species pool should be developed from the nearest possible reference sites, particularly for subordinate species.