This paper evaluates the performance of two numerical wave models, GHOST and STWAVE, with measurements made in an idealized inlet physical model. The emphasis of this paper is on the overall performance of these models in coastal inlets. Both wave models are similar in that they employ a finite-difference method to solve the wave action conservation equation for the steady-state wave spectral transformation. However, these models differ in the computation of diffraction, reflection, wave breaking, and representation of the directional spectrum transformation. The models' performance is compared with a new set of physical model data for four different idealized inlet configurations. Wave height is measured in the physical model by a linear array of capacitance wave gauges, and wave direction is measured by a remote-sensing video-camera system. The comparison with data is presented as mean absolute relative errors of wave height and mean absolute difference of wave direction. Both wave models produced similar results, but neither could accurately describe waves observed in the physical model in inlets and near structures. The mean absolute relative error of wave height prediction from models was between 22 and 40% as compared with the measured data. The mean absolute error of wave direction estimates ranged from 5 to 12 degrees. Overall, wave direction estimates from GHOST in inlets and near structures compared slightly better with measurements.