Floods are a major cause of natural-hazard-related damages and deaths in the U.S. To help mitigate negative flood impacts, hydrometeorological information is used in decisions at a variety of space and time scales. This work examines the interplay among hydrometeorological predictions, decision-making, and flood impacts by analyzing three cases of severe local flooding in the U.S.: the Red River flood of April-May 1997 in Grand Forks, North Dakota; the Fort Collins, Colorado flood in July 1997; and the Pescadero Creek, California flood in February 1998. The three floods occurred under different hydrometeorological and societal circumstances and had different societal impacts, providing an opportunity to compare and contrast lessons learned. Because all three communities had previously experienced significant flooding, the interpretation and use of hydrometeorological predictions in each case were affected by decision-makers' previous experience, and the decisions and outcomes in each case were embedded within longer-term flood risk management contexts. Issues explored include how predictions were used in flood-related decisions, how scientific uncertainty interacted with decision making, how the hydrometeorological community can effectively communicate information to aid flood decision making, and ways that hydrometeorological predictions can (and cannot) help reduce flood losses.