Ensemble average currents from the 15 m depth level of the National Centers for Environmental Prediction's (NCEP) analyses of the tropical Pacific Ocean are evaluated against surface mixed layer current observations obtained from an extensive set of satellite-tracked drifting buoys. These averages display many climatological characteristics of the region but are not intended to serve as a climatology because the data from the analyses are trimmed to match the time-space distribution of the observations. Substantial discrepancies between the analyses and the observations are revealed. First, the near-equatorial meridional currents and divergence have approximately twice the magnitude in the analyses as in the observations. This discrepancy is largely independent of whether temperature profile data are assimilated or not and is attributed to the parameterization of vertical viscosity. Second, the zonal flow in both the North Equatorial Countercurrent (NECC) and the South Equatorial Current (SEC) is much stronger in the analyses than in the observations, especially in the western Pacific. This discrepancy is associated with assimilation of temperature profile data. It arises because salinity is an active variable in the underlying analysis model but is not controlled by boundary fluxes or other observations. Under the uncontrolled influence of advection and strong horizontal diffusion the salinity distribution becomes nearly homogeneous. Consequently, the analyses do not account for observed temperature-salinity correlations when density is computed following assimilation of temperature profile data. This leads to erroneous pressure gradients that drive excessively strong geostrophic currents and force large accelerations near the western boundary. Our results indicate that it is important to consider the consequences on the density structure of neglecting salinity during the assimilation of temperature data. We recommend that surface salinity observations from drifting buoys and volunteer observing ships be initiated to improve the ocean analyses.¿ 1997 American Geophysical Union