An upgraded version of the tidal solutions (FES94.1) is presented, obtained by assimilating an altimeter-derived data set in the finite element hydrodynamic model, following the representer approach. The assimilated data are drawn from the CSR2.0 Texas solutions sampled on a 5¿¿5¿ grid. The assimilation is applied over the Atlantic, Indian, and Pacific Oceans. The standard release of the new FES95.2 solutions is a 0.5¿¿0.5¿ gridded version of the full finite element solutions. The associated tidal prediction model includes 26 constituents. The eight major constituents are drawn directly from the hydrodynamic model: K1, O1, Q1, M2, S2, N2, K2, and 2N2, corrected by assimilation except K2 and 2N2. The other 18 constituents are derived by admittance. Among them are μ2, &ngr;2, L2, T2, M1, P1, J1, and OO1. The quality of these solutions is evaluated by reference to a standard sea truth data set of 95 stations. This quality is significantly improved after the assimilation process is applied: the root-sum-square (RSS) of the differences between solutions and observations, for the eight major constituents, is reduced from 3.8 cm for FES94.1 to 2.8 cm for FES95.2, i.e., a gain of 1 cm. The performances of the prediction model are evaluated by comparing tidal predictions with observations at 59 sites distributed over the world ocean and by looking at the level of variance of the sea surface variability observed by the T/P altimeter at its cross-over track points after tidal correction. These evaluations lead to the same conclusion: this new prediction model performs much better than the one based on FES94.1, because of correction of the major constituents by T/P data assimilation and because of the increase in the number of constituents from 13 to 26. The tidal predictions are at the level of accuracy of those produced by the best recent T/P empirical models. ¿ 1998 American Geophysical Union