A method to predict extreme roof snow loads is described. The procedure utilizes a combination of recent climate data (antecedent snow water equivalent information), probabilistic precipitation amount forecasts and historical extreme snow water equivalent data. Use of the procedure is illustrated using precipitation amount probabilities derived directly from a statistical (MOS) guidance product and using conditional climatological probabilities, both with a lead time of 36 hours. The procedure is applied across a portion of the north-eastern United States where extreme snow loads during January 1996 resulted in widespread roof damage, injuries, and a death. Based on the 90th percentiles of the forecast precipitation amount distributions, the procedure accurately identified the areas in which greater than 100-year return interval snow loads were observed.