A new analysis of the historical temperature and salinity profiles in the tropical Atlantic is done in order to estimate quantitatively the uncertainties in the climatological seasonal variations of 0/400 db dynamic topography. The uncertainties are described by an error covariance matrix which takes into account aliasing and measurement errors, the effect of data gaps and interpolation, as well as the uncertainty of the T-S method that was used to calculate the dynamic height. The standard deviation of the monthly means is found to range between 2 and 10 dyn cm, depending on the data density and the level of eddy activity; substantial error covariances are also introduced by data interpolation. The new data set is used to test objectively the ability of the linear multimode model of Cane (1984), forced by a 20 year wind stress data set, to reproduce the seasonal variations of the dynamic topography. Model-reality intercomparison is done using a multivariate statistical procedure which also takes into account the interannual variability of the forcing, as well as its uncertainties due to random wind stress errors and drag coefficient indeterminacy. The model-reality discrepancies are shown to be too large to be explained by the oceanic and atmospheric uncertainties, and they should be primarily attributed to model shortcomings. Nonetheless, comparison with previous results suggests that the linear model simulates the dynamic topography better than the surface currents; it also reproduces the seasonal variability better than the annual mean. The multimode model works best with the first three vertical modes, although the differences in model performance with two or more vertical modes are not statistically significant.