This study examines the performance of more than ten three-dimensional primitive equation models in hurricane track forecasts. The model forecasts, in addition to some simple statistical models, are used as model guidance for the forecasters at the National Hurricane Center (NHC). The accuracy of these model forecasts is crucial to the official hurricane track prediction by NHC. Over the last 5-10 years the numerical models from various operational prediction centers and research institutes around the world (e.g., NCEP, ECMWF, UKMET office, GFDL, NRL, etc.) have generally been improved with better model physics and increased observations over the oceanic regions. However, how much the model performances in hurricane track forecast have been changed and what are their impacts on the hurricane prediction in general have not been systematicaly investigated. The objective of this study is to determine that how the model track errors vary with the time of forecast periods (i.e., 12, 24, 36, 48, 72 h). The results are compared with the official track forecasts at NHC and the simple statisitcal model estimates. We define the track error as the great circle distance between the forecast positions and the best track positions determined from a post-hurricane evaluation of all available observations. The data used in this study include all named hurricanes from 1990-1997. The forecast errors for each model are averaged yearly for each of the forecast time periods. Our preliminary results show that i) the envelope of the current track forecast errors for all models varies from about 130-178 km at 12 h to 365-520 km at 72 h, ii) the forecast errors has been generally decreased for all time periods over the eight years with the largest improvment in 72 h forecast, and iii) the model errors seem to relate to the number of the storms which vary from year to year and the hurricane track locations which vary from storm to storm. A more comprehensive analysis will be presented at the conference.