Present operational models, based on the primitive equations, frequently fail to produce forecasts of hurricane intensity that are of sufficient accuracy for forecasters. Furthermore, these models lack spatial resolution necessary to represent the small-scale inner core of hurricanes. To overcome such limitations, investigators have turned to nonhydrostatic models designed for higher spatial resolutions than operational models. Recently, the nonhydrostatic version of the PSU/NCAR MM5 has been used to simulate hurricanes at 5 km resolution with success, but the lack of a vortex-following mesh refinement scheme in the available versions of the MM5 have made integrations with 5 km or higher resolution for periods greater than a few hours impractical. In this paper we introduce a vortex-following mesh refinement scheme into MM5 in order to facilitate relatively long (i.e., several days) simulations at 1-2 km horizontal grid spacing (which is the grid spacing necessary to resolve the inner cores of hurricanes). We present results from a set of 6-day simulations of Hurricane Floyd in which we use our modified version of the MM5 with successively refined meshes (down to 1.67 km grid spacing). We compare our simulation results with available observational data sets, including dropsonde temperature and humidity profiles, airborne radar data, TRMM PR and TMI data, and QuikScat scatterometer derived surface wind fields. We find that the forecast track of the model is in good agreement with observation, and that the forecast track is insensitive to model resolution. In contrast, we find that the structure and intensity of the simulated hurricane is quite sensitive to resolution, with the finer-resolution simulations producing stronger and more realistic hurricanes vortices.