Along-track sea level anomalies derived from Geosat altimeter data from November 1986 to November 1988 are assimilated by Kalman filtering into a wind-forced second-baroclinic vertical mode linear model of the tropical Atlantic Ocean. To save computer time, the filter is degraded, mostly by fixing the error covariance matrix of the estimate once the filter has reached its asymptotic behavior. Geosat altimeter data have been processed using improved corrections. The sea surface height variability signal is extracted using the classical along-track technique, relative to a complete reference cycle, and using only tracks longer than 2200 km. This processing has preserved oceanic signals both on large scales (above 1000 km) and on the mesoscale (around 200 km). Sea level anomalies predicted at Principe Island are close to in situ tide gage data, though some differences can be partly related to tidal or orbit error corrections. Oceanographic signals are analyzed from two different sets of fields: one issued from anisotropic space-time objective analysis of Geosat data and the other from the model assimilation. The latter appears as an interesting method to extract low-frequency and propagating signals. Along the equator, eastward propagating features are consistent with Kelvin waves correlated with zonal wind stress anomalies. Upwelling in the Gulf of Guinea is 1 month earlier in 1987 than in 1988. After elimination of the annual and semiannual signals by harmonic analysis, the residual signal over the whole tropical basin, decomposed into complex empirical orthogonal functions, is found dominated by variations between the 2 years, equatorial and tropical signals being anticorrelated.¿ 1997 American Geophysical Union