Hurricane activity varies significantly on an interannual basis making it difficult to develop accurate seasonal predictions of hurricane hazard in the US. This paper uses regression trees to model the relationship between 77 climate variables (e.g. sea surface temperature in various regions, atmospheric humidity, wind shear) and the annual cumulative destructive potential of hurricanes making landfall in the US. This measure of hurricane intensity (calculated from accumulated pressure differences) is used rather than hurricane counts or occurrence rates because hurricane damage is generally modeled as a function of hurricane wind speed. The U.S. coast is divided into three separate regions: (1) the Atlantic coast excluding Florida, (2) Florida, and (3) the Gulf coast excluding Florida, and unique models are developed for each region. The models are developed using all hurricanes that made landfall in the continental United States since 1948. Our analysis demonstrates that regression trees can be used to accurately fit the historical data. The regression tree approach allows for the incorporation of a large number of climate variables, including some which have not been considered in previous hurricane modeling studies. Our study also showed that although there is some agreement between regions 1 and 3 in terms of which climate variables are most important for the modeling hurricane activity, region 2 (Florida) is controlled by a unique set of climate variables.