The Llanos de Ojuelos is an extensive llanura at the southern end of the Chihuahuan Desert. The original composition of the land is mostly grasslands, with interspread shrublands, and small water bodies. Currently, 80% of the area is used for crop production and/or grazing of livestock. As a result, natural habitats have been severely fragmented, and the overall biodiversity affected to an unknown degree. The present work aims at understanding the relationship between rodent assemblages and different types of habitat patches to integrate a landscape approach of small mammal diversity. In the spring of 2008 we inventoried the rodents at 74 locations in the area, through the use of Sherman live-traps (40 traps during two nights, per site; total=5920 night-traps). We covered the major perennial vegetation types. We captured 458 individuals of 20 species of rodents (Perognathus flavus, Chaetodipus hispidus, Chaetodipus nelsoni, C. pennicillatus, Dipodomys ordii, D. merriami, D. phillipsii, Liomys irroratus, Peromyscus boylii, P. difficilis, P. eremicus, P. gratus, P. maniculatus, P. melanophrys, Reithrodontomys fulvescens, R. megalotis, Baiomys taylori, Neotoma goldmani, N. leucodoni, and Onychomys arenarius). We defined a priori 11 rodent habitat categories based on the cover, structure and dominant plant species. Species presence/absences were modeled jointly through a multinomial model, and then used to pool habitats by merging all habitat classes pair-wise and selecting the most parsimonious model by the Akaike Information Criterion for each iteration. This produced 7 habitat classes: Closed arboreal nopalera, mixed nopalera, grassland, leguminous scrubland, Dodonaea scrubland, cultivated nopalera, and oak scrub. Rodent communities analyzed have low levels of organization, with most species thriving in a variety of conditions, although some of them exhibited a certain habitat preference. Communities were distributed according to a gradient from "low open vegetation" to "dense tall vegetation," with fuzzy borders between them. The highest rodent richness was in nopaleras; mid richness, in leguminous scrub communities, and lower richness, in grasslands. This might have resulted from differences in habitat structure and within-habitat heterogeneity. Habitat types seemed also to be structured differently. We are exploring models employing reflectance data to classify satellite images according to habitat classes and to analyze their spatial relationship. Given the patchiness and the low organization level of the communities, biological conservation cannot be carried out in homogeneous areas, but must consider the entire landscape