This paper reviews the status of numerical weather prediction models to forecast the characteristics of mesoscale gravity waves as observed in detailed analyses of surface mesonetwork, wind profiler, and multiple Doppler radar data. The data set consists of multiple published simulations of gravity waves from the 15 December 1987 explosive cyclogenesis event, the 27 March 1994 Palm Sunday tornado outbreak, the 11-12 July 1981 wave event from the Cooperative Convective Precipitation Experiment (CCOPE), the 14 February 1992 event during the STORM-FEST project, and a large-amplitude wave event that occurred during a rapid cyclogenesis event on the East Coast on 4 January 1994. The review shows that gravity waves are ubiquitous features in mesoscale numerical weather prediction models, and that they may dominate quasi-geostrophic signals in the forecast fields. Although current models can predict the wave corridor with reasonable accuracy, they have not shown the ability to make deterministic forecasts of specific wave characteristics (amplitude, wavelength, shape, and speed) with great reliability. The greatest successes have been attained in the two explosive cyclogenesis cases, in which one or two large-amplitude waves formed within a developing dry slot near the inflection axis in the upper-level height field. An efficient wave duct is invariably present downstream of the wave generation region, and unbalanced flow is typically diagnosed in the wave generation region. Ducted wave-CISK modes dominate the types of gravity waves predicted in these mesoscale model studies, yet great uncertainty remains because of the metastable equilibrium with respect to elevated convection found in this environment. Wind profiler and Doppler radar analyses from a very limited subsample of the events appear to support the idea of ducted structures, but detection of these waves prior to the development of wave-induced deep convection is almost non-existent. Shorter and slower waves are predicted as model grid resolution is increased, leading to an increase in the wave spectrum; however, a plethora of waves plagues models run with grid mesh sizes of 3-10 km. The size of the initial disturbance appears to select the scale of the waves. Disturbing sensitivity of simulated gravity wave sensitivities to grid resolution, model initialization time, parameterized shallow and deep convection, and other model physics will be discussed.