Tropical cyclone (TC) forecasts from regional and global numerical weather prediction models have become increasingly more important in recent years as guidance to TC forecasters at both the National Hurricane Center (NHC) and the Joint Typhoon Warning Center (JTWC). Forecasters at NHC routinely utilize the TC track forecasts from the GFDL model, the Navy Operational Global Atmospheric Prediction System (NOGAPS), and the UK Meteorological Office global model (UKMO). Along with these models, forecasters at JTWC also utilize the TC track forecasts from the Japanese typhoon model and global spectral model (JGSM). While the individual track forecasting skill of these different models varies from year to year and basin to basin, their overall skill is quite comparable. The relative independence of the TC track forecasts produced by these models suggests that a simple ensemble average or consensus forecast derived from a combination of these models should be more accurate, on the average, than the forecasts of any of the models individually. The TC forecast performance of a simple ensemble approach utilizing these various models was evaluated for the 1995-96 Atlantic hurricane seasons and for the western North Pacific in 1997. For the Atlantic, the forecast position errors for the ensemble average of the GFDL, NOGAPS, and UKMO forecasts were 65, 105, and 144 nm at 24h, 48h, and 72h, respectively. This represented an error reduction of 16, 20, and 23 percent with respect to the best of the individual models. For the Pacific the forecast position errors for the ensemble average of the NOGAPS, UKMO, and JGSM forecasts were 64, 116, and 174 nm at 24h, 48h, and 72h, respectively, and represented an error reduction of 16, 13, and 12 percent with respect to the best of the individual models. Not only did the ensemble approach produce reduction in the mean forecast position error but it also produced reduction in the spread of the forecast position error. For the Atlantic the 95th percentile of forecast position error was reduced 19, 14, and 23 percent at 24h, 48h, and 72h, respectively, with respect to the best of the individual models. The respective reduction of this quantity for the Pacific was 17, 12, and 17 percent. This reduction in spread could be directly translated to reducing the size of warning areas associated with TC landfall approach forecasts, a major research goal which emerged from the USWRP Hurricane Landfall and Impacts Workshop (17-20 November 1997, Miami FL)