Due to the heavy developments it requires and its significant additional computing cost compared to 3D-Var or reduced rank filters, 4D-Var represents quite an investment. However it offers more flexibility and is based on a more solid theoretical background. NEMOVAR, the global variational assimilation system for NEMO has been used operationally both at the Met Office and ECMWF for some time in its 3D-Fgat formulation. Incremental 4D-Var has now been developed in this framework. This presentation will compare the relative merits of 3D-and 4D-var over a 1 year test case mimicking the ECMWF ORA-S4 global 1° ocean reanalysis. In particular one will discuss the different possible minimisation strategies (assimilation window length, multiple outer loops, IAU, continuous DA, ...) and their impact on the analysis quality. Finally, if time permits, preliminary results using a multi-resolution approach in order to reduce the computational burden will be presented.