The Weather Research and Forecasting (WRF) Modeling System is a cooperative program among scientists at NCAR, FSL, OU, NCEP, AFWA and elsewhere to develop the next generation multiscale prediction system. As part of this prediction system, a three-dimensional variational (3D-VAR) assimilation system is being developed. Capitalizing upon the collective experience of their organizations, the participating scientists have designed a 3D-VAR system and begun its construction. The system is being coded to both run on serial processors and to exploit massively parallel features available on a variety of larger computing platforms. The 3D-VAR system has two major goals: (1) to be able to analyze information from a wide range of conventional sources, together with satellite radiances, radar radial velocities and reflectivities and other remotely-sensed data and (2) is to accommodate a geographical and flow dependence of scale and variance parameters of the background error covariances together with a specification of three-dimensional anisotropy in a general way. The design and coding pays attention to the different requirements of the intended users, who will include research meteorologists and operational forecasting centers. The design, current configuration and future enhancements to the assimilation system will be presented.