Several methods are tested to reduce the undesirable gravity wave oscillations in numerical integrations of NRL's Limited Area Weather Prediction model. The split-explicit temporal integration scheme used in the model is shown by itself to reduce the amplitude of the external gravity wave oscillations in the first six hours of integration. In a static non-linear initialization procedure, a diagnostic relation is derived for the geopotential on the sigma surfaces of the model. The procedure provides a balanced initial state, except for an initial adjustment of the vertical motion in the first five hours of integration. A vertical mode initialization procedure is also developed for the model. The scheme is tested on two grids, of differing domain size and grid resolution. Convergence of the scheme is shown to be faster for the case of smoother topography along the lateral boundaries. Integrations with two different lateral boundary treatments are compared. The procedure is shown to prevent gravity wave oscillations without producing a mean drift in the surface pressure, to provide a balanced mass divergence, and to produce smaller changes to the initial mass and wind fields, compared to the static initialization. Keywords: Atmosphere models; Mathematical filters; Weather forecasting; Filtering gravity waves; Static non-linear mass balance.