In the framework of the Argo program, profiling drifting floats are now routinely launched in the world's oceans. These floats provide (among other information) data about their position, sampled every ten days, representative of their Lagrangian drift. Previous work (Nodet,2006) have shown the interest of assimilating this new type of data. The assimilation of Lagrangian-type data such as position of drifting floats is not straightforward; it involves the careful implementation of a complex, non-linear observation operator but it is of importance for an application in operational oceanography. We propose here to study the addition of such observations in an operational variational system: the ocean model NEMO, coupled with the incremental 4D-Var tool NEMOVAR. In a physical configuration, we compare the impact of several variational assimiliation strategies, for a realistic set of observations. The impact of the incremental 4D-Var algorithm is compared to previous 3DFGAT experiments; Lagrangian observations contribute to improve the statistical quality of the analyzed state, but their impact can be limited, since the assimilation of other observations may introduce disorder in the velocity.