Two mesoscale numerical weather prediction models, Eta and the Rapid Update Cycle (RUC), are currently being run operationally at the U.S. National Weather Service’s National Meteorological Center to produce nationwide weather forecasts. Improvements in weather forecast accuracy depend on increasing model resolution, which is limited by available computing resources. Massively Parallel Processing (MPP) offers a cost-effective way of increasing computing resources. At the Forecast Systems Laboratory, we are developing parallel versions of both models that can be easily ported between different MPP systems and traditional sequential machines. To support this effort, we have developed the Nearest Neighbor Tool (NNT), a software library of high level routines that greatly reduces the effort required to parallelize finite difference approximation weather forecast models 1. Here, we describe our experiences using NNT to parallelize the Eta and RUC models, discuss performance optimization strategies, and present performance results on various MPP systems. Performance is evaluated in three areas: computation, inter-process communication, and I/O. Most source code optimizations are targeted toward reducing inter-process communication time. NNT provides the user with a flexible method of minimizing total execution time by trading communication time for redundant computations. Finally, we describe our progress toward demonstrating the feasibility of using MPP systems in an operational environment. I/O performance and machine stability and reliability are also critical to meeting operational time lines. * Also affiliated with Science and Technology Corporation, Hampton, VA † Joint collaboration with Cooperative Institute for Research in the Atmosphere, Colorado State Univ., Ft. Collins, CO