ABSTRACTA neural network-based model for interior longwave radiative heat transfer has been developed and implemented into a new computer code, BERHT (Building Energy with Radiative Heat Transfer). The model accounts for the non-gray effect of absorbing species in a building environment and the geometric effect of a three-dimensional building structure. Numerical studies have been carried out on a rectangular single-story building. For nominal concentration of CO2, H2O, and small particulates, results show that the effect of radiative heat transfer is important. The surface emissivity of enclosure walls and optical properties of the absorbing/emitting medium are demonstrated to have significant effects on the distribution of heat transfer between convection and radiation, as well as the transient behavior of the indoor air temperature. Supplemental studies provide an insight that the one-zone, well-mixed model used in building energy simulation generates a “fictitious” non-local heat transfer behavior, leading to uncertainties in the understanding of the radiative heat transfer effect.