The impact of assimilating Argo data into an initial field on the short-term forecasting accuracy of temperature and salinity is quantitatively estimated by using a forecasting system of the western North Pacific, on the base of the Princeton ocean model with a generalized coordinate system (POMgcs). This system uses a sequential multigrid three-dimensional variational (3DVAR) analysis scheme to assimilate observation data. Two numerical experiments were conducted with and without Argo temperature and salinity profile data besides conventional temperature and salinity profile data and sea surface height anomaly (SSHa) and sea surface temperature (SST) in the process of assimilating data into the initial fields. The forecast errors are estimated by using independent temperature and salinity profiles during the forecasting period, including the vertical distributions of the horizontally averaged root mean square errors (H-RMSEs) and the horizontal distributions of the vertically averaged mean errors (MEs) and the temporal variation of spatially averaged root mean square errors (S-RMSEs). Comparison between the two experiments shows that the assimilation of Argo data significantly improves the forecast accuracy, with 24% reduction of H-RMSE maximum for the temperature, and the salinity forecasts are improved more obviously, averagely dropping of 50% for H-RMSEs in depth shallower than 300 m. Such improvement is caused by relatively uniform sampling of both temperature and salinity from the Argo drifters in time and space.