The ecological benefits of changing cattle grazing practices in the western United States remain controversial, due in part to a lack of experimentation. In 1997 we initiated an experimental study of two rangeland alternatives, cattle removal and high-impact grazing, and compared grassland community responses with those with more conventional, moderate grazing practices. The study was conducted in a high-elevation, semiarid grassland near Flagstaff, Arizona (U.S.A.). We conducted annual plant surveys of modified Whittaker plots for 8 years and examined plant composition shifts among treatments and years. High-impact grazing had strong directional effects that led to a decline in perennial forb cover and an increase in annual plants, particularly the exotic cheatgrass (Bromus tectorum L.). A twofold increase in plant cover by exotic species followed a severe drought in the sixth year of the study, and this increase was greatest in the high-impact grazing plots, where native cover declined by one-half. Cattle removal resulted in little increase in native plant cover and reduced plant species richness relative to the moderate grazing control. Our results suggest that some intermediate level of cattle grazing may maintain greater levels of native plant diversity than the alternatives of cattle removal or high-density, short-duration grazing practices. Furthermore, episodic drought interacts with cattle grazing, leading to infrequent, but biologically important shifts in plant communities. Our results demonstrate the importance of climatic variation in determining ecological effects of grazing practices, and we recommend improving conservation efforts in arid rangelands by developing management plans that anticipate this variation.