A global ocean model has been developed within the context of TOPEX/Poseidon (T/P) on the basis of a finite element hydrodynamic modeling approach combined with data assimilation based on the representer method. The solution produced at global scale represents a spectacular improvement over what was available before the era of T/P. Typically, the mean discrepancy between the main tidal components and a hundred in situ tide gauges is 1.7 cm for M2 and 1 cm or less for the other components. However, the accuracies of this solution and of that produced by different authors within the T/P tide working group are all worse near coastlines and over continental shelves. This is the case for our finite element solution (FES) FES94.1 solution over the Yellow Sea and the East China Sea (YS-ECS) where the discrepancy is 33 cm for the M2 tide and 9 cm for the K1 tide when compared to a set of 192 tide gauges distributed along the coastlines. This is due largely to the complex geometry of the basin and the limited knowledge of the bathymetry. The aim of this paper is to investigate how our FES model can be improved with a dedicated application focused on one of the most energetic coastal basins: the YS-ECS (~180 Gigawatts for M2, i.e. 8% of the global energy dissipated in the whole ocean). For this application, a finite grid was implemented with a resolution down to 5 km along the coasts and over the continental shelf break. Particular attention was paid to bathymetry by complementing the ETOP05 database with regional maps. Sensitivity tests to the tuning of the bottom friction and the nonlinear interactions between the diurnal and semidiurnal components allow us to investigate the impact of dissipation parameterization and to produce an optimal solution, without any data assimilation, for nine main components of the tidal spectrum (improvement by a factor of 2). M2 distance to the observations is now ~17.5 cm (global variance of 92 cm), and K1 is 4 cm (global variance of 21 cm). As nonlinear components are significant over these basins, the two main quarter-diurnal components are calculated. The improvements are explained by five points: (1) the choice of a mixed friction, (2) a refinement of the mesh, (3) a choice of reliable boundary conditions, (4) a refined topography and, (5) the use of a specific friction coefficient 1.5¿10-3. Surprisingly, the energy budget for the M2 component leads to a dissipation similar to the value estimated by the global FES94.1. This is a major result of this study, which leads us to the conclusion that when forcing a regional model with sea level boundary conditions along the open limits, the simulated velocity field adjusts to the tuning of friction coefficients in order to dissipate the same amount of energy made available at its open boundaries. ¿ 2000 American Geophysical Union