The unexpected weakening of Hurricane Debby on August 23, 2000 is examined using in-situ observations, satellite data, and NCEP global analyses. Additionally, output from the operational GFDL (2000 version) is compared with the observed data and select reruns from the modified GFDL (2003 version). The 2003 version includes improved horizontal and vertical resolution, modified cumulus parameterization and boundary layer closure schemes, a modified pressure gradient computation, a new mass initialization scheme, and improved model initialization and analysis. Comparisons between the 2000 and 2003 versions of the GFDL output reveal some positive impacts of these upgrades on the model’s performance in the Debby case. A poleward track and high intensity bias observed during the operational 2000 runs are due to excessive downstream ridge development which created a weaker shear environment and stronger southerly steering flow. Although these tendencies are not completely removed in the 2003 reruns, they are reduced and have yielded improved model forecasts for the analyzed cases during Debby