Recently the NCEP's regional four-dimensional variational (4DVAR) data assimilation system has been updated to accommodate the need for operational implementation. Few important changes are:

1. minimization space is defined in coarser resolution,
   
2. system can employ a combination of fine and coarse resolution forecast models, and corresponding adjoints,
   
3. only the nonlinear models are used in forward integration, and
   
4. background and model error covariances are anisotropic and non-homogeneous.
   

The performance of a new 4DVAR system will be shown.

Key words: variational, four-dimensional, data assimilation, regional