In this paper, we introduce a predictive, physics-based model, i.e., the so-called tilted-mirror model (tm-model), for optical modeling of rough rear surfaces on silicon solar cells. An enhanced method of using transfer matrices at the rear-side interface of solar cells is developed and combined with Monte Carlo ray tracing. As a result, a physically consistent and precise simulation of the spectral reflectance is achieved, thus leading to a predictive quality of the simulations that could previously not be reached for solar cells with a remaining irregular rear-surface roughness. This advance in optical simulation enables the researcher to directly analyze the effects of varying rear-side passivation materials and thicknesses, as well as the impact of different surface morphologies on the gained charge-carrier generation rate of a solar cell. A comparison with the Phong model shows that the tm-model is able to simulate the generated photocurrent Jph more accurately, as it is shown that the Phong model tends to overestimate this value due to imprecise calculation of charge-carrier generation. In an application of the tm-model to passivated emitter and rear cells, it is shown that a strong planarization of the rear surface leads to an improvement in photogenerated current up to 0.13 mA/cm2 compared with a weak planarization.