An ensemble of simulations of a coupled chemistry-climate model is completed for 1960--2100. The simulations are divided into two periods, 1960--2005 and 1990--2100. The modeled total ozone amount decrease throughout the atmosphere from the 1960s until about 2000--2005, depending on latitude. The Antarctic ozone hole develops rapidly in the model from about the late 1970s, in agreement with observations, but it does not disappear until about 2065, about 15 years later than previous estimates. Spring averaged ozone takes even longer to recover to 1980 values. Ozone amounts in the Antarctic are determined largely by halogen amounts. In contrast, in the Arctic, ozone recovers to 1980 values about 25--35 years earlier, depending on the recovery criterion adopted. By the end of the 21st century, the climate change associated with greenhouse gas changes gives rise to a significant superrecovery of ozone in the Arctic but a less marked recovery in the Antarctic. For both polar regions, ensemble and interannual variability is greater in the future than in the past, and hence the timing of the full recovery of polar ozone is very sensitive to the definition of recovery. It is suggested that the range of recovery rates between the hemispheres simulated in the model is related to the overall increase in the strength of the Brewer-Dobson circulation, driven by increases in greenhouse gas concentrations.