Abstract:GRAPES_TCM2.6(GT2.6), the updated version of the numerical model G RAPES_TCM2.1（GT2.1）, is used to post forecast tracks of TCs occurred over the Northw estern Pacific and the South China Sea during 2005. A series of detailed perform ance verifications for the numerical prediction of tropical cyclone tracks is pr esented. The results show that GT2.6 has a goodperformance on TC track pr ediction. The total mean forecast errors (MFE) at 24, 48 and 72h are 135.8, 230. 7and 336.0km, respectively. The result also shows that the MFE at 12h is somewha t large (about 100km). The significant improvement of GT2.6's track prediction f or the turning TC can be obtained through the modification of the forecast track s with systematic bias. The model is found to have very good forecast skill for the cases that occurred in the west of 130E, especially for those close to the s outheast of China. In general, the forecast track of GT2.6 is faster (slower ) b efore (after) 48 hours than the optimal track, and the speed bias at 12 hour is larger during the forecast period (approximately 1m/s). In comparison with GT2. 1, GT2.6 has similar performance stability and MFE characteristics on 0 48h TC t rack prediction, except that the 48h forecast of GT 2.6 is significantly better than that of GT 2.1.  Primary results indicate that several essential measures should be taken to furt her improve the track forecast skill of GT2.6. They are replacing the vortex-env ironment separation scheme with a stricter one, adding persistence vector to imp rove the initial motion and an asymmetric bogus vortex in the initial time, as w ell as enlarging the forecast domain and properly setting its position. 