We live-trapped rodents in 2000-2001 at eight sites on a 3160 ha grassland and mesquite-oak savanna in southeastern Arizona that had been ungrazed since 1968, and on eight paired sites on adjacent cattle ranches. There were 917 captures of 14 species during 5760 trap-nights. Four species of Muridae (Sigmodon fulviventer, Baiomys taylori, Reithrodontomys megalotis and R. fulvescens) were significantly more common on ungrazed plots, while no species was more abundant on grazed plots. However, Heteromyidae as a group (especially Chaetodipus hispidus and Perognathus flavus) comprised a significantly higher proportion of total captures on grazed plots, and heteromyids as a percentage of total captures was positively correlated across all plots with amount of bare ground. One of the eight cross-fence sites also had been trapped in 1981-1983. In the 17 y between trapping events at this site: (1) the grass canopy on both grazed and ungrazed plots had become dominated by taller species, (2) a kangaroo rat (Dipodomys merriami) that had been the second most common species in grazed areas disappeared from both plots, (3) pocket mice increased on the grazed plot and declined on the ungrazed plot and (4) Muridae (excluding Peromyscus) as a percent of all captures increased by greater than 1.5-fold on both plots. Based on these results, and those from other field studies, we propose a model for the composition of rodent communities in grass/shrublands of the Southwest and Intermountain West, based on ground cover. Kangaroo rats (Dipodomys spp.) are abundant in areas with the most bare soil, Muridae (specifically, Sigmodon, Baiomys and Reithrodontomys) dominate areas with the most and tallest ground cover, and pocket mice (Chaetodipus and Perognathus) are common in areas of intermediate cover. In relatively mesic grasslands, livestock grazing and fire drive the rodent community toward one dominated by heteromyids instead of murids. In more arid landscapes, grazing and fire favor kangaroo rats over pocket mice.