Woody plant encroachment into grassy ecosystems is a worldwide phenomenon that radically transforms ecosystem services. Billions of dollars have been spent to remove woody plants, but there is great uncertainty about the conditions in which such removals are beneficial. We conducted a collaborative, large-scale monitoring experiment in the Chihuahuan Desert of southern New Mexico, USA, to test whether woody plant removal restored historical states, created novel states, or exacerbated land degradation and to examine the environmental conditions affecting outcomes. We monitored vegetation cover in 43 pairs of plots representing herbicide-treated and herbicide-untreated conditions of the same plant community and environmental setting, including measurements at baseline, 5, 10, and in some cases 15 years posttreatment. We compared outcomes to measurements in reference sites. Woody plant removal led to increases in plant species richness and perennial grass cover, but increases were due to disturbance-adapted grasses rather than species characteristic of reference states. On average, grass cover in treatments did not attain levels observed in the reference state. Negative effects of woody plant removal on total canopy cover (related to soil erosion risk) and other plant functional groups of concern for wildlife were not observed. Elevation, slope, and baseline cover were important predictors of treatment-associated gains in plant cover, while higher grazing intensity was related to increases in richness and forb cover. Our results indicate that while woody plant removal cannot be considered to restore a reference state due to the continued absence of reference grass species, it does not lead to obvious land degradation. Furthermore, more positive outcomes are maximized where (and when) soil moisture limitations are lowest or the cover of responding plants is highest prior to treatments. We recommend that future restoration actions be conducted as experiments that pay special attention to co-production mechanisms, standardized monitoring methods, and salient and easily measured indicators.