Two sets of monthly sea surface wind stress over the tropical Atlantic Ocean are compared. The datasets are based on the ECMWF analyses during 1980–87 (the E-winds) and the monthly pseudo-wind stress from ship observations for the same period (the S-winds). Our examination shows that both datasets give qualitatively similar mean fields and annual cycles. Quantitatively, the zonal component of the E-winds is larger than that of the S-winds, especially in the winter hemisphere. The strongest southeast trades of the E-winds are also shifted to the east of the strongest southeast trades of the S-winds. In the vicinity of the ITCZ, the E-winds are more zonally oriented so that the convergence zone is not as clearly defined. Interannually, both datasets show that the northeast trades were gradually strengthening from 1980 to 1986. The southeast trade winds, on the other hand, were anomalously strong during 1981–83, but weak during 1984–86. With the E-winds, the southeast trades decreased gradually during 1981–84, and with the S-winds, the southeast trades are maintained until late 1983, followed by a rapid weakening. In comparison with the E-winds, the S-winds interannual fluctuations over the central and eastern part of the tropical south and equatorial Atlantic are weak. The sensitivity of an ocean general circulation model to the uncertainty of surface wind forcing as exemplified by these two datasets is examined. It is found that the systematic errors in the mean state and annual cycle of the model simulated sea surface temperature (SST) and upper ocean heat content (HC) are not sensitive to the differences in wind forcings. On the other hand, significantly different fluctuations of both the SSTs and the HCs on interannual timescales are generated by the simulations forced with the two wind data, respectively. A comparison between the observed and simulated SST anomalies shows that both simulations are reasonably close to the observations in the tropical north Atlantic Ocean. In the tropical south Atlantic, the E-winds produce a better simulation of the SST anomalies. Especially, the gradual weakening of the E-winds during 1981–84 produces an SST tendency consistent with observations, which are not shown in the S-winds simulation. However, the E-winds anomalies are poor during 1980–81 as judged by the comparison between the simulated and the observed SST anomalies.