The French MERCATOR project is developing several operational ocean forecasting systems to take part in the Global Ocean Data Assimilation Experiment (GODAE). Prototype systems are designed to simulate (1) the Atlantic and Mediterranean Sea (from 1/3o to 1/15o ), and (2) the global ocean circulation (from 2o to 1/4o ). The first generation assimilation scheme referred to as SAM1 has been implemented in the operational system. It provides routine weekly analyses and forecasts. SAM1 includes an altimetry-only version (SAM1-v1), and a fully multivariate version (SAM1-v2) permitting to assimilate vertical profiles and SST in addition to altimetry (JASON, ERS-2 and GFO). The SAM1 scheme is based on the SOFA reduced order interpolation scheme (LEGOS, Toulouse). It uses vertical/horizontal separation of error statistics, and order reduction in the vertical in terms of multivariate Empirical Orthogonal Functions (EOFs) of temperature: T, salinity: S, and barotropic streamfunction: ψ . The next generation assimilation system referred as SAM2 is being developed from the SEEK (Singular Evolutive Extended Kalman) algorithm (LEGI, Grenoble). This scheme is a Reduced Order Kalman Filter using a 3D multivariate modal decomposition of the forecast error covariance as well as an adaptive scheme to specify parameters of the forecast error. The use of the SEEK filter and its 3D modal representation for the error statistic is intended to overcome some of the limitations of SAM1 in highly inhomogeneous, anisotropic, and nonseparable regions of the world ocean such as shallow areas, as well as in the surface layer. A second objective for SAM2 will be to consistently propagate error estimates between successive assimilation cycles. We also developed several methods in order to generate 3D EOF basis both from local or global EOF calculation. Comparisons between these various multivariate approaches (SAM1 and SAM2) will be presented and discussed.