A barotropic spectral model (BSM) is developed to investigate the possibility of forecasting tropical cyclone tracks with global, general circulation models. The model is governed by a barotropic vorticity equation in spherical coordinates which is solved wing a spectral method with spherical harmonic basis functions. The model was run with a triangular truncation of 128 on half of the northern hemisphere (180°W–0°W), and was initialized using horizontal winds from the NMC analyses vertically averaged from 1000 to 100 mb. The storm circulation is represented by a specified axisymmetric vortex and the model was tested by making 30 track forecasts of Atlantic tropical cyclones (13 storms) which occurred from 1979 to 1984. The skill of the model was assessed by comparing the track forecasts to forecasts from a model based on climatology and persistence (CLIPER). The BSM has statistically significant skill for 24 and 36 h track forecasts and longer range of skill for forecasts of low-latitude storms. For low-latitude storms, the BSM had than the operational SANBAR and moveable fine mesh (MFM) models. The sensitivity of the model to the horizontal resolution is tested. These results suggest that track forecasts could be made with a global spectral model with a triangular truncation of about 96. It might then be feasible to make track forecasts with a global spectral model similar to the operational model at the European Centre for Medium Range Weather Forecasts which uses a triangular truncation of 106. The sensitivity of the model to the domain sin and to the specification of the initial vortex is also investigated. These results show that when simple lateral boundary conditions are used, the track forecast errors rapidly increase when the model domain is made smaller than half of a hemisphere. These results also show that the track errors are very insensitive to the size of the vortex, provided that the vortex is not unrealistically large. When the shape of the vortex profile is changed to include an anticylonic circulation at large radii, the track errors are smallest when the total angular momentum of the vortex is close to zero. The errors rapidly increase as the total angular momentum becomes negative. The effect of modifying the initial analyses so that the deep-layer mean wind in the storm region is consistent with the previous storm motion is also studied. The track errors show the most reduction when the analyses within a radius of about 1000 km of the norm am modified.