A method for determining a severe weather index based on the observed seasonal cycle of tornados is introduced and applied to the North American Regional Reanalysis. This index based on convective precipitation and vertical shear, exhibits skill at reproducing the spatial distribution of tornado risk and the interannual variability of the tornado season from 1979-2001. The index is then applied to a pair of high-resolution climate simulations (T159 and T1279) in order to examine their ability to reproduce the observations of the 20th century and to compare their projections for the 21st century. The results show that the higher resolution model better reproduces the spatial distribution and interannual variability of tornado risk. Both simulations indicate an increased risk of tornados in a warmer world, consistent with previous studies. The lower resolution model projects dire changes, exceeding 25%, over much of the eastern U.S. However, the higher resolution model shows more moderate changes and its greater skill at reproducing the observations may suggest that these climate change projections are more credible. This difference is due to larger changes in the mean atmospheric circulation projected by the lower resolution model. A poleward shift of the mid-latitude westerly winds is a common feature of climate change simulations, but the implication for the distributions of tornados has not been previously noted. This study underscores the fact that the magnitude and seasonality of such a change in circulation will have consequences for our weather patterns. If regional projections of climate change are going to be useful for policy makers, the fidelity of these subtle tendencies must be examined closely.