Abstract: A statistical model of waterspout probability was developed for wet-season (June–September) days over the Florida Keys. An analysis was performed on over 200 separate variables derived from Key West 1200 UTC daily wet-season soundings during the period 2006–14. These variables were separated into two subsets: days on which a waterspout was reported anywhere in the Florida Keys coastal waters and days on which no waterspouts were reported. Days on which waterspouts were reported were determined from the National Weather Service (NWS) Key West local storm reports. The sounding at Key West was used for this analysis since it was assumed to be representative of the atmospheric environment over the area evaluated in this study. The probability of a waterspout report day was modeled using multiple logistic regression with selected predictors obtained from the sounding variables. The final model containing eight separate variables was validated using repeated fivefold cross validation, and its performance was compared to that of an existing waterspout index used as a benchmark. The performance of the model was further validated in forecast mode using an independent verification wet-season dataset from 2015–16 that was not used to define or train the model. The eight-predictor model was found to produce a probability forecast with robust skill relative to climatology and superior to the benchmark waterspout index in both the cross validation and in the independent verification.