Although the short-term forecasts of Hurricane Katrina were about as good as one could hope for, the present state of the art has somewhat larger average track and intensity errors than was the case with Katrina. Moreover, forecasts with lead times greater than about 3 days were highly uncertain; and Katrina did not even exist as a named storm 6 days before it struck New Orleans. In this talk, I will discuss prospects for improved hurricane forecasting, focusing on longer term forecasts and on prediction of storm strength, which has lagged behind improvements in storm track prediction. I will argue that improvements in hurricane prediction will depend not just on better models but on intelligent design of adaptive measurement systems and far more serious attention to measurement of the upper ocean in advance of the storms. The potential value of large ensembles of model forecasts will also be emphasized.