Abstract We present an investigation of the azimuthal bimodality of the wind-wave spectrum for waves shorter than the dominant scale comparing numerical model solutions of developing waves from idealized experiments using WAVEWATCH III (WW3). The wave solutions were forced with the “exact” Webb–Resio–Tracy (WRT) nonlinear energy fluxes and the direct interaction approximation (DIA) with three different combinations of wind input and breaking dissipation parameterizations. The WRT gives larger azimuthal bimodal amplitudes compared to the DIA regardless of wind input/dissipation. The widely used wind input/dissipation parameterizations (i.e., ST4 and ST6) generally give narrow directional distributions with relatively small bimodal amplitudes and lobe separations compared to field measurements. These biases are significantly improved by the breaking dissipation of Romero (R2019). Moreover, the ratio of the resolved cross- to downwind mean square slope is significantly lower for ST4 and ST6 compared to R2019. The overlap integral relevant for the prediction of microseisms is several orders of magnitude smaller for ST4 and ST6 compared to R2019, which nearly agrees with a semiempirical model. Significance Statement Spectral gravity wave models generally cannot accurately predict the directional distribution which impacts their ability to predict the resolved down- and crosswind mean square slopes and the generation of microseisms. Our analysis shows that a directionally narrow spectral energy dissipation, accounting for long-wave–short-wave modulation, can significantly improve the directional distribution of the wind-wave spectrum by coupling to the nonlinear energy fluxes due to wave–wave interactions, which has important implications for improved predictions of the mean square slopes and the generation of microseisms.