Ecosystems that supported multiple biological communities during the Holocene and that form alternative stable states pose challenges for identifying appropriate and feasible target communities for ecological restoration and conservation strategies for climate adaptation. In the Chihuahuan Desert, shrubland and perennial grassland alternatively dominated for millennia. However, grassland declined beginning 100–150 years ago during concomitant change in both climate and anthropogenic land use, making it unclear whether restoring grassland is appropriate or feasible in the emerging climate. In Guadalupe Mountains National Park, Texas, USA, we characterized remnant grasslands and performed experiments including outplanting and shrub reduction to assess feasibility and techniques for restoring desert grassland. In remnant patches, grass cover was more temporally variable than shrub cover, as grass cover varied 1.5× between years. Survival of outplanted grasses, either under Larrea tridentata shrubs or in the open with or without woody material encircling outplants for protection, was high (95 %) after seven months but declined to 2 % after 19 months. Response of perennial grass cover was variable after shrub and litter removal on soil surfaces varying in stoniness. Results suggest that 1) restoration can increase native desert grasses but gains may be intermittent; 2) reducing shrubs to facilitate natural expansion of remnant grasses and propagules may be more effective than outplanting grasses; and 3) restoring native grassland by reducing shrubs appears feasible even in dry conditions projected for the emerging climate. Findings that native grasses can increase at least intermittently suggest that practitioners have multiple candidate communities for conservation management and climate change adaptation as grassland, shrubland, and mixed communities each provide unique habitat.