A classification scheme is proposed as an improvement to the widely used U.S. Department of Agriculture and Arnold Arboretum plant hardiness maps using maximum and minimum air temperature, precipitation, wind speed, sunshine, relative humidity, growing season length, temperature extremes, growth units and elevation. For this analysis, the U.S.A. and Canada were divided into 1234 grid boxes for which monthly averages of each of the climatological variables were calculated. To reduce the size of the original 136 variable data set and eliminate the intercorrelation which existed among these variables, principal component analysis was used and resulted in the formation of a three-component data set. These components were subjected to the flexible clustering technique to isolate regions which, based on the components, were similar, with respect to winter plant hardiness and survival. Twenty-three distinct hardiness clusters with sound climatological and horticultural bases were produced by the clustering procedure. The clusters defined regions with similar vegetative distributions and ecological significance. In the east, clusters were oriented latitudinally and resembled the USDA zones. A distinct cluster encompassing the southern Appalachian Mountains was also formed. In the central U.S.A., the orientation of the clusters was influenced by both temperature and precipitation. The clusters were irregular in mountainous regions and became longitudinally oriented along the Pacific coast. Several of the western clusters encompassed such climatically unique regions as the desert southwest, Great Basin and Rocky Mountains.